program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.6.1"}, {"coremlc-version", "3304.7.1"}})]
{
    func main<ios16>(tensor<int32, [1, 128]> attention_mask, tensor<int32, [1, 128]> input_ids) {
            tensor<int32, []> var_10 = const()[name = tensor<string, []>("op_10"), val = tensor<int32, []>(-1)];
            tensor<bool, []> var_23 = const()[name = tensor<string, []>("op_23"), val = tensor<bool, []>(true)];
            tensor<int32, []> hidden_states_1_axis_0 = const()[name = tensor<string, []>("hidden_states_1_axis_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [32003, 2048]> model_model_embed_tokens_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32771072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32771200))), name = tensor<string, []>("model_model_embed_tokens_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([32003, 2048])];
            tensor<int32, []> hidden_states_1_cast_batch_dims_0 = const()[name = tensor<string, []>("hidden_states_1_cast_batch_dims_0"), val = tensor<int32, []>(0)];
            tensor<fp16, [1, 128, 2048]> hidden_states_1_cast = gather(axis = hidden_states_1_axis_0, batch_dims = hidden_states_1_cast_batch_dims_0, indices = input_ids, x = model_model_embed_tokens_weight_to_fp16_palettized)[name = tensor<string, []>("hidden_states_1_cast")];
            tensor<int32, [1]> var_108_axes_0 = const()[name = tensor<string, []>("op_108_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1, 128]> var_108 = expand_dims(axes = var_108_axes_0, x = attention_mask)[name = tensor<string, []>("op_108")];
            tensor<int32, [1]> var_109_axes_0 = const()[name = tensor<string, []>("op_109_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1, 1, 1, 128]> var_109 = expand_dims(axes = var_109_axes_0, x = var_108)[name = tensor<string, []>("op_109")];
            tensor<int32, [4]> var_112_reps_0 = const()[name = tensor<string, []>("op_112_reps_0"), val = tensor<int32, [4]>([1, 1, 128, 1])];
            tensor<int32, [1, 1, 128, 128]> var_112 = tile(reps = var_112_reps_0, x = var_109)[name = tensor<string, []>("op_112")];
            tensor<fp16, []> var_16_to_fp16 = const()[name = tensor<string, []>("op_16_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<string, []> var_113_to_fp16_dtype_0 = const()[name = tensor<string, []>("op_113_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 1, 128, 128]> cast_2 = cast(dtype = var_113_to_fp16_dtype_0, x = var_112)[name = tensor<string, []>("cast_2")];
            tensor<fp16, [1, 1, 128, 128]> inverted_mask_cast = sub(x = var_16_to_fp16, y = cast_2)[name = tensor<string, []>("inverted_mask_cast")];
            tensor<string, []> var_115_dtype_0 = const()[name = tensor<string, []>("op_115_dtype_0"), val = tensor<string, []>("bool")];
            tensor<fp16, []> var_11_to_fp16 = const()[name = tensor<string, []>("op_11_to_fp16"), val = tensor<fp16, []>(-inf)];
            tensor<bool, [1, 1, 128, 128]> cast_1 = cast(dtype = var_115_dtype_0, x = inverted_mask_cast)[name = tensor<string, []>("cast_1")];
            tensor<fp16, [1, 1, 128, 128]> var_116_cast = select(a = var_11_to_fp16, b = inverted_mask_cast, cond = cast_1)[name = tensor<string, []>("op_116_cast")];
            tensor<fp16, [1, 1, 128, 128]> op_97_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [8192]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32771328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32779584))), name = tensor<string, []>("op_97_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 128, 128])];
            tensor<fp16, [1, 1, 128, 128]> attention_mask_cast = add(x = var_116_cast, y = op_97_to_fp16_palettized)[name = tensor<string, []>("attention_mask_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16 = const()[name = tensor<string, []>("op_13_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_125_cast = pow(x = hidden_states_1_cast, y = var_13_promoted_to_fp16)[name = tensor<string, []>("op_125_cast")];
            tensor<int32, [1]> var_126 = const()[name = tensor<string, []>("op_126"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_1_cast = reduce_mean(axes = var_126, keep_dims = var_23, x = var_125_cast)[name = tensor<string, []>("variance_1_cast")];
            tensor<fp16, []> var_128_to_fp16 = const()[name = tensor<string, []>("op_128_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_129_cast = add(x = variance_1_cast, y = var_128_to_fp16)[name = tensor<string, []>("op_129_cast")];
            tensor<fp16, []> var_130_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_130_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_130_cast = rsqrt(epsilon = var_130_epsilon_0_to_fp16, x = var_129_cast)[name = tensor<string, []>("op_130_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_5_cast = mul(x = hidden_states_1_cast, y = var_130_cast)[name = tensor<string, []>("hidden_states_5_cast")];
            tensor<fp16, [2048]> model_model_layers_0_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_0_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32779712)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_9_cast = mul(x = model_model_layers_0_input_layernorm_weight_to_fp16, y = hidden_states_5_cast)[name = tensor<string, []>("hidden_states_9_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_0_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32783872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34881088))), name = tensor<string, []>("model_model_layers_0_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [2048]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34881216)))];
            tensor<fp16, [1, 128, 2048]> linear_0_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_0_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("linear_0_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_0_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34885376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35147584))), name = tensor<string, []>("model_model_layers_0_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [256]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35147712)))];
            tensor<fp16, [1, 128, 256]> linear_1_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_0_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("linear_1_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_0_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35148288))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35410496))), name = tensor<string, []>("model_model_layers_0_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_2_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_0_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_9_cast)[name = tensor<string, []>("linear_2_cast")];
            tensor<int32, [4]> var_150 = const()[name = tensor<string, []>("op_150"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_151_cast = reshape(shape = var_150, x = linear_0_cast)[name = tensor<string, []>("op_151_cast")];
            tensor<int32, [4]> q_1_perm_0 = const()[name = tensor<string, []>("q_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_153 = const()[name = tensor<string, []>("op_153"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_154_cast = reshape(shape = var_153, x = linear_1_cast)[name = tensor<string, []>("op_154_cast")];
            tensor<int32, [4]> key_states_3_perm_0 = const()[name = tensor<string, []>("key_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_156 = const()[name = tensor<string, []>("op_156"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_157_cast = reshape(shape = var_156, x = linear_2_cast)[name = tensor<string, []>("op_157_cast")];
            tensor<int32, [4]> hidden_states_15_perm_0 = const()[name = tensor<string, []>("hidden_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 1, 128, 64]> cos_1_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35410624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35414784))), name = tensor<string, []>("cos_1_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> transpose_109 = transpose(perm = q_1_perm_0, x = var_151_cast)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [1, 32, 128, 64]> var_183_cast = mul(x = transpose_109, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_183_cast")];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = tensor<string, []>("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = tensor<string, []>("x1_1_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = tensor<string, []>("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_1_cast = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = transpose_109)[name = tensor<string, []>("x1_1_cast")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = tensor<string, []>("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = tensor<string, []>("x2_1_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = tensor<string, []>("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_1_cast = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = transpose_109)[name = tensor<string, []>("x2_1_cast")];
            tensor<fp16, []> const_11_promoted_to_fp16 = const()[name = tensor<string, []>("const_11_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_194_cast = mul(x = x2_1_cast, y = const_11_promoted_to_fp16)[name = tensor<string, []>("op_194_cast")];
            tensor<bool, []> var_196_interleave_0 = const()[name = tensor<string, []>("op_196_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_196_cast = concat(axis = var_10, interleave = var_196_interleave_0, values = (var_194_cast, x1_1_cast))[name = tensor<string, []>("op_196_cast")];
            tensor<fp16, [1, 1, 128, 64]> sin_1_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35414912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35419072))), name = tensor<string, []>("sin_1_to_fp16_palettized"), shape = tensor<uint32, [4]>([1, 1, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> var_197_cast = mul(x = var_196_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_197_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_3_cast = add(x = var_183_cast, y = var_197_cast)[name = tensor<string, []>("query_states_3_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_108 = transpose(perm = key_states_3_perm_0, x = var_154_cast)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, 4, 128, 64]> var_199_cast = mul(x = transpose_108, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_199_cast")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = tensor<string, []>("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = tensor<string, []>("x1_3_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = tensor<string, []>("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_3_cast = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = transpose_108)[name = tensor<string, []>("x1_3_cast")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = tensor<string, []>("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = tensor<string, []>("x2_3_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = tensor<string, []>("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_3_cast = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = transpose_108)[name = tensor<string, []>("x2_3_cast")];
            tensor<fp16, []> const_14_promoted_to_fp16 = const()[name = tensor<string, []>("const_14_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_210_cast = mul(x = x2_3_cast, y = const_14_promoted_to_fp16)[name = tensor<string, []>("op_210_cast")];
            tensor<bool, []> var_212_interleave_0 = const()[name = tensor<string, []>("op_212_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_212_cast = concat(axis = var_10, interleave = var_212_interleave_0, values = (var_210_cast, x1_3_cast))[name = tensor<string, []>("op_212_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_213_cast = mul(x = var_212_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_213_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_11_cast = add(x = var_199_cast, y = var_213_cast)[name = tensor<string, []>("hidden_states_11_cast")];
            tensor<int32, [1]> var_222_axes_0 = const()[name = tensor<string, []>("op_222_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_222_cast = expand_dims(axes = var_222_axes_0, x = hidden_states_11_cast)[name = tensor<string, []>("op_222_cast")];
            tensor<int32, [5]> hidden_states_13_reps_0 = const()[name = tensor<string, []>("hidden_states_13_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_13_cast = tile(reps = hidden_states_13_reps_0, x = var_222_cast)[name = tensor<string, []>("hidden_states_13_cast")];
            tensor<int32, [4]> var_230 = const()[name = tensor<string, []>("op_230"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_5_cast = reshape(shape = var_230, x = hidden_states_13_cast)[name = tensor<string, []>("key_states_5_cast")];
            tensor<int32, [1]> var_239_axes_0 = const()[name = tensor<string, []>("op_239_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_107 = transpose(perm = hidden_states_15_perm_0, x = var_157_cast)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_239_cast = expand_dims(axes = var_239_axes_0, x = transpose_107)[name = tensor<string, []>("op_239_cast")];
            tensor<int32, [5]> hidden_states_17_reps_0 = const()[name = tensor<string, []>("hidden_states_17_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_17_cast = tile(reps = hidden_states_17_reps_0, x = var_239_cast)[name = tensor<string, []>("hidden_states_17_cast")];
            tensor<int32, [4]> var_247 = const()[name = tensor<string, []>("op_247"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_3_cast = reshape(shape = var_247, x = hidden_states_17_cast)[name = tensor<string, []>("value_states_3_cast")];
            tensor<int32, [4]> var_249_perm_0 = const()[name = tensor<string, []>("op_249_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_250_transpose_x_0 = const()[name = tensor<string, []>("op_250_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_250_transpose_y_0 = const()[name = tensor<string, []>("op_250_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_106 = transpose(perm = var_249_perm_0, x = key_states_5_cast)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [1, 32, 128, 128]> var_250_cast = matmul(transpose_x = var_250_transpose_x_0, transpose_y = var_250_transpose_y_0, x = query_states_3_cast, y = transpose_106)[name = tensor<string, []>("op_250_cast")];
            tensor<fp16, []> _inversed_attn_weights_1_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_1_cast = mul(x = var_250_cast, y = _inversed_attn_weights_1_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_1_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_1_cast = add(x = _inversed_attn_weights_1_cast, y = attention_mask_cast)[name = tensor<string, []>("input_1_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_254_cast = softmax(axis = var_10, x = input_1_cast)[name = tensor<string, []>("op_254_cast")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_1_cast = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = var_254_cast, y = value_states_3_cast)[name = tensor<string, []>("attn_output_1_cast")];
            tensor<int32, [4]> var_257_perm_0 = const()[name = tensor<string, []>("op_257_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_259 = const()[name = tensor<string, []>("op_259"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_105 = transpose(perm = var_257_perm_0, x = attn_output_1_cast)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 128, 2048]> input_3_cast = reshape(shape = var_259, x = transpose_105)[name = tensor<string, []>("input_3_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_0_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35419200))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37516416))), name = tensor<string, []>("model_model_layers_0_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_3_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_0_self_attn_o_proj_weight_to_fp16_palettized, x = input_3_cast)[name = tensor<string, []>("linear_3_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_21_cast = add(x = hidden_states_1_cast, y = linear_3_cast)[name = tensor<string, []>("hidden_states_21_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_1 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_1"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_266_cast = pow(x = hidden_states_21_cast, y = var_13_promoted_to_fp16_1)[name = tensor<string, []>("op_266_cast")];
            tensor<int32, [1]> var_267 = const()[name = tensor<string, []>("op_267"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_3_cast = reduce_mean(axes = var_267, keep_dims = var_23, x = var_266_cast)[name = tensor<string, []>("variance_3_cast")];
            tensor<fp16, []> var_269_to_fp16 = const()[name = tensor<string, []>("op_269_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_270_cast = add(x = variance_3_cast, y = var_269_to_fp16)[name = tensor<string, []>("op_270_cast")];
            tensor<fp16, []> var_271_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_271_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_271_cast = rsqrt(epsilon = var_271_epsilon_0_to_fp16, x = var_270_cast)[name = tensor<string, []>("op_271_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_25_cast = mul(x = hidden_states_21_cast, y = var_271_cast)[name = tensor<string, []>("hidden_states_25_cast")];
            tensor<fp16, [2048]> model_model_layers_0_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_0_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37516544)))];
            tensor<fp16, [1, 128, 2048]> input_5_cast = mul(x = model_model_layers_0_post_attention_layernorm_weight_to_fp16, y = hidden_states_25_cast)[name = tensor<string, []>("input_5_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_0_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37520704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43287936))), name = tensor<string, []>("model_model_layers_0_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [5632]> linear_4_bias_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2816]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43288064))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43290944))), name = tensor<string, []>("linear_4_bias_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([5632])];
            tensor<fp16, [1, 128, 5632]> linear_4_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_0_mlp_gate_proj_weight_to_fp16_palettized, x = input_5_cast)[name = tensor<string, []>("linear_4_cast")];
            tensor<fp16, [1, 128, 5632]> var_283_cast = silu(x = linear_4_cast)[name = tensor<string, []>("op_283_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_0_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43291072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49058304))), name = tensor<string, []>("model_model_layers_0_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_5_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_0_mlp_up_proj_weight_to_fp16_palettized, x = input_5_cast)[name = tensor<string, []>("linear_5_cast")];
            tensor<fp16, [1, 128, 5632]> input_9_cast = mul(x = var_283_cast, y = linear_5_cast)[name = tensor<string, []>("input_9_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_0_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49058432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54825664))), name = tensor<string, []>("model_model_layers_0_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_6_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_0_mlp_down_proj_weight_to_fp16_palettized, x = input_9_cast)[name = tensor<string, []>("linear_6_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_31_cast = add(x = hidden_states_21_cast, y = linear_6_cast)[name = tensor<string, []>("hidden_states_31_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_2 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_2"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_296_cast = pow(x = hidden_states_31_cast, y = var_13_promoted_to_fp16_2)[name = tensor<string, []>("op_296_cast")];
            tensor<int32, [1]> var_297 = const()[name = tensor<string, []>("op_297"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_5_cast = reduce_mean(axes = var_297, keep_dims = var_23, x = var_296_cast)[name = tensor<string, []>("variance_5_cast")];
            tensor<fp16, []> var_299_to_fp16 = const()[name = tensor<string, []>("op_299_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_300_cast = add(x = variance_5_cast, y = var_299_to_fp16)[name = tensor<string, []>("op_300_cast")];
            tensor<fp16, []> var_301_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_301_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_301_cast = rsqrt(epsilon = var_301_epsilon_0_to_fp16, x = var_300_cast)[name = tensor<string, []>("op_301_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_35_cast = mul(x = hidden_states_31_cast, y = var_301_cast)[name = tensor<string, []>("hidden_states_35_cast")];
            tensor<fp16, [2048]> model_model_layers_1_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_1_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54825792)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_39_cast = mul(x = model_model_layers_1_input_layernorm_weight_to_fp16, y = hidden_states_35_cast)[name = tensor<string, []>("hidden_states_39_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_1_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54829952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56927168))), name = tensor<string, []>("model_model_layers_1_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_7_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_1_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_39_cast)[name = tensor<string, []>("linear_7_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_1_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56927296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57189504))), name = tensor<string, []>("model_model_layers_1_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_8_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_1_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_39_cast)[name = tensor<string, []>("linear_8_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_1_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57189632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57451840))), name = tensor<string, []>("model_model_layers_1_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_9_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_1_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_39_cast)[name = tensor<string, []>("linear_9_cast")];
            tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_322_cast = reshape(shape = var_321, x = linear_7_cast)[name = tensor<string, []>("op_322_cast")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = tensor<string, []>("q_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_324 = const()[name = tensor<string, []>("op_324"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_325_cast = reshape(shape = var_324, x = linear_8_cast)[name = tensor<string, []>("op_325_cast")];
            tensor<int32, [4]> key_states_9_perm_0 = const()[name = tensor<string, []>("key_states_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_328_cast = reshape(shape = var_327, x = linear_9_cast)[name = tensor<string, []>("op_328_cast")];
            tensor<int32, [4]> hidden_states_45_perm_0 = const()[name = tensor<string, []>("hidden_states_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_104 = transpose(perm = q_3_perm_0, x = var_322_cast)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 32, 128, 64]> var_354_cast = mul(x = transpose_104, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_354_cast")];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = tensor<string, []>("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = tensor<string, []>("x1_5_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = tensor<string, []>("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_5_cast = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = transpose_104)[name = tensor<string, []>("x1_5_cast")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = tensor<string, []>("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = tensor<string, []>("x2_5_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = tensor<string, []>("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_5_cast = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = transpose_104)[name = tensor<string, []>("x2_5_cast")];
            tensor<fp16, []> const_28_promoted_to_fp16 = const()[name = tensor<string, []>("const_28_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_365_cast = mul(x = x2_5_cast, y = const_28_promoted_to_fp16)[name = tensor<string, []>("op_365_cast")];
            tensor<bool, []> var_367_interleave_0 = const()[name = tensor<string, []>("op_367_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_367_cast = concat(axis = var_10, interleave = var_367_interleave_0, values = (var_365_cast, x1_5_cast))[name = tensor<string, []>("op_367_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_368_cast = mul(x = var_367_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_368_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_7_cast = add(x = var_354_cast, y = var_368_cast)[name = tensor<string, []>("query_states_7_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_103 = transpose(perm = key_states_9_perm_0, x = var_325_cast)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 4, 128, 64]> var_370_cast = mul(x = transpose_103, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_370_cast")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = tensor<string, []>("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = tensor<string, []>("x1_7_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = tensor<string, []>("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_7_cast = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = transpose_103)[name = tensor<string, []>("x1_7_cast")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = tensor<string, []>("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = tensor<string, []>("x2_7_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = tensor<string, []>("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_7_cast = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = transpose_103)[name = tensor<string, []>("x2_7_cast")];
            tensor<fp16, []> const_31_promoted_to_fp16 = const()[name = tensor<string, []>("const_31_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_381_cast = mul(x = x2_7_cast, y = const_31_promoted_to_fp16)[name = tensor<string, []>("op_381_cast")];
            tensor<bool, []> var_383_interleave_0 = const()[name = tensor<string, []>("op_383_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_383_cast = concat(axis = var_10, interleave = var_383_interleave_0, values = (var_381_cast, x1_7_cast))[name = tensor<string, []>("op_383_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_384_cast = mul(x = var_383_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_384_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_41_cast = add(x = var_370_cast, y = var_384_cast)[name = tensor<string, []>("hidden_states_41_cast")];
            tensor<int32, [1]> var_393_axes_0 = const()[name = tensor<string, []>("op_393_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_393_cast = expand_dims(axes = var_393_axes_0, x = hidden_states_41_cast)[name = tensor<string, []>("op_393_cast")];
            tensor<int32, [5]> hidden_states_43_reps_0 = const()[name = tensor<string, []>("hidden_states_43_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_43_cast = tile(reps = hidden_states_43_reps_0, x = var_393_cast)[name = tensor<string, []>("hidden_states_43_cast")];
            tensor<int32, [4]> var_401 = const()[name = tensor<string, []>("op_401"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_11_cast = reshape(shape = var_401, x = hidden_states_43_cast)[name = tensor<string, []>("key_states_11_cast")];
            tensor<int32, [1]> var_410_axes_0 = const()[name = tensor<string, []>("op_410_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_102 = transpose(perm = hidden_states_45_perm_0, x = var_328_cast)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_410_cast = expand_dims(axes = var_410_axes_0, x = transpose_102)[name = tensor<string, []>("op_410_cast")];
            tensor<int32, [5]> hidden_states_47_reps_0 = const()[name = tensor<string, []>("hidden_states_47_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_47_cast = tile(reps = hidden_states_47_reps_0, x = var_410_cast)[name = tensor<string, []>("hidden_states_47_cast")];
            tensor<int32, [4]> var_418 = const()[name = tensor<string, []>("op_418"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_7_cast = reshape(shape = var_418, x = hidden_states_47_cast)[name = tensor<string, []>("value_states_7_cast")];
            tensor<int32, [4]> var_420_perm_0 = const()[name = tensor<string, []>("op_420_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_421_transpose_x_0 = const()[name = tensor<string, []>("op_421_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_421_transpose_y_0 = const()[name = tensor<string, []>("op_421_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_101 = transpose(perm = var_420_perm_0, x = key_states_11_cast)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 32, 128, 128]> var_421_cast = matmul(transpose_x = var_421_transpose_x_0, transpose_y = var_421_transpose_y_0, x = query_states_7_cast, y = transpose_101)[name = tensor<string, []>("op_421_cast")];
            tensor<fp16, []> _inversed_attn_weights_5_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_5_cast = mul(x = var_421_cast, y = _inversed_attn_weights_5_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_5_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_11_cast = add(x = _inversed_attn_weights_5_cast, y = attention_mask_cast)[name = tensor<string, []>("input_11_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_425_cast = softmax(axis = var_10, x = input_11_cast)[name = tensor<string, []>("op_425_cast")];
            tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_5_cast = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = var_425_cast, y = value_states_7_cast)[name = tensor<string, []>("attn_output_5_cast")];
            tensor<int32, [4]> var_428_perm_0 = const()[name = tensor<string, []>("op_428_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_430 = const()[name = tensor<string, []>("op_430"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_100 = transpose(perm = var_428_perm_0, x = attn_output_5_cast)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 128, 2048]> input_13_cast = reshape(shape = var_430, x = transpose_100)[name = tensor<string, []>("input_13_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_1_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57451968))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59549184))), name = tensor<string, []>("model_model_layers_1_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_10_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_1_self_attn_o_proj_weight_to_fp16_palettized, x = input_13_cast)[name = tensor<string, []>("linear_10_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_51_cast = add(x = hidden_states_31_cast, y = linear_10_cast)[name = tensor<string, []>("hidden_states_51_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_3 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_3"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_437_cast = pow(x = hidden_states_51_cast, y = var_13_promoted_to_fp16_3)[name = tensor<string, []>("op_437_cast")];
            tensor<int32, [1]> var_438 = const()[name = tensor<string, []>("op_438"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_7_cast = reduce_mean(axes = var_438, keep_dims = var_23, x = var_437_cast)[name = tensor<string, []>("variance_7_cast")];
            tensor<fp16, []> var_440_to_fp16 = const()[name = tensor<string, []>("op_440_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_441_cast = add(x = variance_7_cast, y = var_440_to_fp16)[name = tensor<string, []>("op_441_cast")];
            tensor<fp16, []> var_442_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_442_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_442_cast = rsqrt(epsilon = var_442_epsilon_0_to_fp16, x = var_441_cast)[name = tensor<string, []>("op_442_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_55_cast = mul(x = hidden_states_51_cast, y = var_442_cast)[name = tensor<string, []>("hidden_states_55_cast")];
            tensor<fp16, [2048]> model_model_layers_1_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_1_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59549312)))];
            tensor<fp16, [1, 128, 2048]> input_15_cast = mul(x = model_model_layers_1_post_attention_layernorm_weight_to_fp16, y = hidden_states_55_cast)[name = tensor<string, []>("input_15_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_1_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59553472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65320704))), name = tensor<string, []>("model_model_layers_1_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_11_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_1_mlp_gate_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("linear_11_cast")];
            tensor<fp16, [1, 128, 5632]> var_454_cast = silu(x = linear_11_cast)[name = tensor<string, []>("op_454_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_1_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65320832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71088064))), name = tensor<string, []>("model_model_layers_1_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_12_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_1_mlp_up_proj_weight_to_fp16_palettized, x = input_15_cast)[name = tensor<string, []>("linear_12_cast")];
            tensor<fp16, [1, 128, 5632]> input_19_cast = mul(x = var_454_cast, y = linear_12_cast)[name = tensor<string, []>("input_19_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_1_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71088192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76855424))), name = tensor<string, []>("model_model_layers_1_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_13_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_1_mlp_down_proj_weight_to_fp16_palettized, x = input_19_cast)[name = tensor<string, []>("linear_13_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_61_cast = add(x = hidden_states_51_cast, y = linear_13_cast)[name = tensor<string, []>("hidden_states_61_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_4 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_4"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_467_cast = pow(x = hidden_states_61_cast, y = var_13_promoted_to_fp16_4)[name = tensor<string, []>("op_467_cast")];
            tensor<int32, [1]> var_468 = const()[name = tensor<string, []>("op_468"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_9_cast = reduce_mean(axes = var_468, keep_dims = var_23, x = var_467_cast)[name = tensor<string, []>("variance_9_cast")];
            tensor<fp16, []> var_470_to_fp16 = const()[name = tensor<string, []>("op_470_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_471_cast = add(x = variance_9_cast, y = var_470_to_fp16)[name = tensor<string, []>("op_471_cast")];
            tensor<fp16, []> var_472_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_472_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_472_cast = rsqrt(epsilon = var_472_epsilon_0_to_fp16, x = var_471_cast)[name = tensor<string, []>("op_472_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_65_cast = mul(x = hidden_states_61_cast, y = var_472_cast)[name = tensor<string, []>("hidden_states_65_cast")];
            tensor<fp16, [2048]> model_model_layers_2_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_2_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76855552)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_69_cast = mul(x = model_model_layers_2_input_layernorm_weight_to_fp16, y = hidden_states_65_cast)[name = tensor<string, []>("hidden_states_69_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_2_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76859712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78956928))), name = tensor<string, []>("model_model_layers_2_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_14_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_2_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_69_cast)[name = tensor<string, []>("linear_14_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_2_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78957056))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79219264))), name = tensor<string, []>("model_model_layers_2_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_15_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_2_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_69_cast)[name = tensor<string, []>("linear_15_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_2_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79219392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79481600))), name = tensor<string, []>("model_model_layers_2_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_16_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_2_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_69_cast)[name = tensor<string, []>("linear_16_cast")];
            tensor<int32, [4]> var_492 = const()[name = tensor<string, []>("op_492"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_493_cast = reshape(shape = var_492, x = linear_14_cast)[name = tensor<string, []>("op_493_cast")];
            tensor<int32, [4]> q_5_perm_0 = const()[name = tensor<string, []>("q_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_496_cast = reshape(shape = var_495, x = linear_15_cast)[name = tensor<string, []>("op_496_cast")];
            tensor<int32, [4]> key_states_15_perm_0 = const()[name = tensor<string, []>("key_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_498 = const()[name = tensor<string, []>("op_498"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_499_cast = reshape(shape = var_498, x = linear_16_cast)[name = tensor<string, []>("op_499_cast")];
            tensor<int32, [4]> hidden_states_75_perm_0 = const()[name = tensor<string, []>("hidden_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_99 = transpose(perm = q_5_perm_0, x = var_493_cast)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 32, 128, 64]> var_525_cast = mul(x = transpose_99, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_525_cast")];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = tensor<string, []>("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = tensor<string, []>("x1_9_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = tensor<string, []>("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_9_cast = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = transpose_99)[name = tensor<string, []>("x1_9_cast")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = tensor<string, []>("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = tensor<string, []>("x2_9_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = tensor<string, []>("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_9_cast = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = transpose_99)[name = tensor<string, []>("x2_9_cast")];
            tensor<fp16, []> const_45_promoted_to_fp16 = const()[name = tensor<string, []>("const_45_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_536_cast = mul(x = x2_9_cast, y = const_45_promoted_to_fp16)[name = tensor<string, []>("op_536_cast")];
            tensor<bool, []> var_538_interleave_0 = const()[name = tensor<string, []>("op_538_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_538_cast = concat(axis = var_10, interleave = var_538_interleave_0, values = (var_536_cast, x1_9_cast))[name = tensor<string, []>("op_538_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_539_cast = mul(x = var_538_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_539_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_11_cast = add(x = var_525_cast, y = var_539_cast)[name = tensor<string, []>("query_states_11_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_98 = transpose(perm = key_states_15_perm_0, x = var_496_cast)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 4, 128, 64]> var_541_cast = mul(x = transpose_98, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_541_cast")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = tensor<string, []>("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = tensor<string, []>("x1_11_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = tensor<string, []>("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_11_cast = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = transpose_98)[name = tensor<string, []>("x1_11_cast")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = tensor<string, []>("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = tensor<string, []>("x2_11_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = tensor<string, []>("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_11_cast = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = transpose_98)[name = tensor<string, []>("x2_11_cast")];
            tensor<fp16, []> const_48_promoted_to_fp16 = const()[name = tensor<string, []>("const_48_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_552_cast = mul(x = x2_11_cast, y = const_48_promoted_to_fp16)[name = tensor<string, []>("op_552_cast")];
            tensor<bool, []> var_554_interleave_0 = const()[name = tensor<string, []>("op_554_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_554_cast = concat(axis = var_10, interleave = var_554_interleave_0, values = (var_552_cast, x1_11_cast))[name = tensor<string, []>("op_554_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_555_cast = mul(x = var_554_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_555_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_71_cast = add(x = var_541_cast, y = var_555_cast)[name = tensor<string, []>("hidden_states_71_cast")];
            tensor<int32, [1]> var_564_axes_0 = const()[name = tensor<string, []>("op_564_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_564_cast = expand_dims(axes = var_564_axes_0, x = hidden_states_71_cast)[name = tensor<string, []>("op_564_cast")];
            tensor<int32, [5]> hidden_states_73_reps_0 = const()[name = tensor<string, []>("hidden_states_73_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_73_cast = tile(reps = hidden_states_73_reps_0, x = var_564_cast)[name = tensor<string, []>("hidden_states_73_cast")];
            tensor<int32, [4]> var_572 = const()[name = tensor<string, []>("op_572"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_17_cast = reshape(shape = var_572, x = hidden_states_73_cast)[name = tensor<string, []>("key_states_17_cast")];
            tensor<int32, [1]> var_581_axes_0 = const()[name = tensor<string, []>("op_581_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_97 = transpose(perm = hidden_states_75_perm_0, x = var_499_cast)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_581_cast = expand_dims(axes = var_581_axes_0, x = transpose_97)[name = tensor<string, []>("op_581_cast")];
            tensor<int32, [5]> hidden_states_77_reps_0 = const()[name = tensor<string, []>("hidden_states_77_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_77_cast = tile(reps = hidden_states_77_reps_0, x = var_581_cast)[name = tensor<string, []>("hidden_states_77_cast")];
            tensor<int32, [4]> var_589 = const()[name = tensor<string, []>("op_589"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_11_cast = reshape(shape = var_589, x = hidden_states_77_cast)[name = tensor<string, []>("value_states_11_cast")];
            tensor<int32, [4]> var_591_perm_0 = const()[name = tensor<string, []>("op_591_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_592_transpose_x_0 = const()[name = tensor<string, []>("op_592_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_592_transpose_y_0 = const()[name = tensor<string, []>("op_592_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_96 = transpose(perm = var_591_perm_0, x = key_states_17_cast)[name = tensor<string, []>("transpose_96")];
            tensor<fp16, [1, 32, 128, 128]> var_592_cast = matmul(transpose_x = var_592_transpose_x_0, transpose_y = var_592_transpose_y_0, x = query_states_11_cast, y = transpose_96)[name = tensor<string, []>("op_592_cast")];
            tensor<fp16, []> _inversed_attn_weights_9_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_9_cast = mul(x = var_592_cast, y = _inversed_attn_weights_9_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_9_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_21_cast = add(x = _inversed_attn_weights_9_cast, y = attention_mask_cast)[name = tensor<string, []>("input_21_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_596_cast = softmax(axis = var_10, x = input_21_cast)[name = tensor<string, []>("op_596_cast")];
            tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_9_cast = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = var_596_cast, y = value_states_11_cast)[name = tensor<string, []>("attn_output_9_cast")];
            tensor<int32, [4]> var_599_perm_0 = const()[name = tensor<string, []>("op_599_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_601 = const()[name = tensor<string, []>("op_601"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_95 = transpose(perm = var_599_perm_0, x = attn_output_9_cast)[name = tensor<string, []>("transpose_95")];
            tensor<fp16, [1, 128, 2048]> input_23_cast = reshape(shape = var_601, x = transpose_95)[name = tensor<string, []>("input_23_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_2_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79481728))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81578944))), name = tensor<string, []>("model_model_layers_2_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_17_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_2_self_attn_o_proj_weight_to_fp16_palettized, x = input_23_cast)[name = tensor<string, []>("linear_17_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_81_cast = add(x = hidden_states_61_cast, y = linear_17_cast)[name = tensor<string, []>("hidden_states_81_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_5 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_5"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_608_cast = pow(x = hidden_states_81_cast, y = var_13_promoted_to_fp16_5)[name = tensor<string, []>("op_608_cast")];
            tensor<int32, [1]> var_609 = const()[name = tensor<string, []>("op_609"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_11_cast = reduce_mean(axes = var_609, keep_dims = var_23, x = var_608_cast)[name = tensor<string, []>("variance_11_cast")];
            tensor<fp16, []> var_611_to_fp16 = const()[name = tensor<string, []>("op_611_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_612_cast = add(x = variance_11_cast, y = var_611_to_fp16)[name = tensor<string, []>("op_612_cast")];
            tensor<fp16, []> var_613_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_613_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_613_cast = rsqrt(epsilon = var_613_epsilon_0_to_fp16, x = var_612_cast)[name = tensor<string, []>("op_613_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_85_cast = mul(x = hidden_states_81_cast, y = var_613_cast)[name = tensor<string, []>("hidden_states_85_cast")];
            tensor<fp16, [2048]> model_model_layers_2_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_2_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81579072)))];
            tensor<fp16, [1, 128, 2048]> input_25_cast = mul(x = model_model_layers_2_post_attention_layernorm_weight_to_fp16, y = hidden_states_85_cast)[name = tensor<string, []>("input_25_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_2_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81583232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87350464))), name = tensor<string, []>("model_model_layers_2_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_18_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_2_mlp_gate_proj_weight_to_fp16_palettized, x = input_25_cast)[name = tensor<string, []>("linear_18_cast")];
            tensor<fp16, [1, 128, 5632]> var_625_cast = silu(x = linear_18_cast)[name = tensor<string, []>("op_625_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_2_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87350592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93117824))), name = tensor<string, []>("model_model_layers_2_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_19_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_2_mlp_up_proj_weight_to_fp16_palettized, x = input_25_cast)[name = tensor<string, []>("linear_19_cast")];
            tensor<fp16, [1, 128, 5632]> input_29_cast = mul(x = var_625_cast, y = linear_19_cast)[name = tensor<string, []>("input_29_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_2_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93117952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98885184))), name = tensor<string, []>("model_model_layers_2_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_20_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_2_mlp_down_proj_weight_to_fp16_palettized, x = input_29_cast)[name = tensor<string, []>("linear_20_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_91_cast = add(x = hidden_states_81_cast, y = linear_20_cast)[name = tensor<string, []>("hidden_states_91_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_6 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_6"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_638_cast = pow(x = hidden_states_91_cast, y = var_13_promoted_to_fp16_6)[name = tensor<string, []>("op_638_cast")];
            tensor<int32, [1]> var_639 = const()[name = tensor<string, []>("op_639"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_13_cast = reduce_mean(axes = var_639, keep_dims = var_23, x = var_638_cast)[name = tensor<string, []>("variance_13_cast")];
            tensor<fp16, []> var_641_to_fp16 = const()[name = tensor<string, []>("op_641_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_642_cast = add(x = variance_13_cast, y = var_641_to_fp16)[name = tensor<string, []>("op_642_cast")];
            tensor<fp16, []> var_643_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_643_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_643_cast = rsqrt(epsilon = var_643_epsilon_0_to_fp16, x = var_642_cast)[name = tensor<string, []>("op_643_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_95_cast = mul(x = hidden_states_91_cast, y = var_643_cast)[name = tensor<string, []>("hidden_states_95_cast")];
            tensor<fp16, [2048]> model_model_layers_3_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_3_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98885312)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_99_cast = mul(x = model_model_layers_3_input_layernorm_weight_to_fp16, y = hidden_states_95_cast)[name = tensor<string, []>("hidden_states_99_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_3_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98889472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100986688))), name = tensor<string, []>("model_model_layers_3_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_21_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_3_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("linear_21_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_3_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100986816))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101249024))), name = tensor<string, []>("model_model_layers_3_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_22_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_3_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("linear_22_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_3_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101249152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101511360))), name = tensor<string, []>("model_model_layers_3_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_23_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_3_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_99_cast)[name = tensor<string, []>("linear_23_cast")];
            tensor<int32, [4]> var_663 = const()[name = tensor<string, []>("op_663"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_664_cast = reshape(shape = var_663, x = linear_21_cast)[name = tensor<string, []>("op_664_cast")];
            tensor<int32, [4]> q_7_perm_0 = const()[name = tensor<string, []>("q_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_666 = const()[name = tensor<string, []>("op_666"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_667_cast = reshape(shape = var_666, x = linear_22_cast)[name = tensor<string, []>("op_667_cast")];
            tensor<int32, [4]> key_states_21_perm_0 = const()[name = tensor<string, []>("key_states_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_669 = const()[name = tensor<string, []>("op_669"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_670_cast = reshape(shape = var_669, x = linear_23_cast)[name = tensor<string, []>("op_670_cast")];
            tensor<int32, [4]> hidden_states_105_perm_0 = const()[name = tensor<string, []>("hidden_states_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_94 = transpose(perm = q_7_perm_0, x = var_664_cast)[name = tensor<string, []>("transpose_94")];
            tensor<fp16, [1, 32, 128, 64]> var_696_cast = mul(x = transpose_94, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_696_cast")];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = tensor<string, []>("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = tensor<string, []>("x1_13_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = tensor<string, []>("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_13_cast = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = transpose_94)[name = tensor<string, []>("x1_13_cast")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = tensor<string, []>("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = tensor<string, []>("x2_13_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = tensor<string, []>("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_13_cast = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = transpose_94)[name = tensor<string, []>("x2_13_cast")];
            tensor<fp16, []> const_62_promoted_to_fp16 = const()[name = tensor<string, []>("const_62_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_707_cast = mul(x = x2_13_cast, y = const_62_promoted_to_fp16)[name = tensor<string, []>("op_707_cast")];
            tensor<bool, []> var_709_interleave_0 = const()[name = tensor<string, []>("op_709_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_709_cast = concat(axis = var_10, interleave = var_709_interleave_0, values = (var_707_cast, x1_13_cast))[name = tensor<string, []>("op_709_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_710_cast = mul(x = var_709_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_710_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_15_cast = add(x = var_696_cast, y = var_710_cast)[name = tensor<string, []>("query_states_15_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_93 = transpose(perm = key_states_21_perm_0, x = var_667_cast)[name = tensor<string, []>("transpose_93")];
            tensor<fp16, [1, 4, 128, 64]> var_712_cast = mul(x = transpose_93, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_712_cast")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = tensor<string, []>("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = tensor<string, []>("x1_15_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = tensor<string, []>("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_15_cast = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = transpose_93)[name = tensor<string, []>("x1_15_cast")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = tensor<string, []>("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = tensor<string, []>("x2_15_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = tensor<string, []>("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_15_cast = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = transpose_93)[name = tensor<string, []>("x2_15_cast")];
            tensor<fp16, []> const_65_promoted_to_fp16 = const()[name = tensor<string, []>("const_65_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_723_cast = mul(x = x2_15_cast, y = const_65_promoted_to_fp16)[name = tensor<string, []>("op_723_cast")];
            tensor<bool, []> var_725_interleave_0 = const()[name = tensor<string, []>("op_725_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_725_cast = concat(axis = var_10, interleave = var_725_interleave_0, values = (var_723_cast, x1_15_cast))[name = tensor<string, []>("op_725_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_726_cast = mul(x = var_725_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_726_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_101_cast = add(x = var_712_cast, y = var_726_cast)[name = tensor<string, []>("hidden_states_101_cast")];
            tensor<int32, [1]> var_735_axes_0 = const()[name = tensor<string, []>("op_735_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_735_cast = expand_dims(axes = var_735_axes_0, x = hidden_states_101_cast)[name = tensor<string, []>("op_735_cast")];
            tensor<int32, [5]> hidden_states_103_reps_0 = const()[name = tensor<string, []>("hidden_states_103_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_103_cast = tile(reps = hidden_states_103_reps_0, x = var_735_cast)[name = tensor<string, []>("hidden_states_103_cast")];
            tensor<int32, [4]> var_743 = const()[name = tensor<string, []>("op_743"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_23_cast = reshape(shape = var_743, x = hidden_states_103_cast)[name = tensor<string, []>("key_states_23_cast")];
            tensor<int32, [1]> var_752_axes_0 = const()[name = tensor<string, []>("op_752_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_92 = transpose(perm = hidden_states_105_perm_0, x = var_670_cast)[name = tensor<string, []>("transpose_92")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_752_cast = expand_dims(axes = var_752_axes_0, x = transpose_92)[name = tensor<string, []>("op_752_cast")];
            tensor<int32, [5]> hidden_states_107_reps_0 = const()[name = tensor<string, []>("hidden_states_107_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_107_cast = tile(reps = hidden_states_107_reps_0, x = var_752_cast)[name = tensor<string, []>("hidden_states_107_cast")];
            tensor<int32, [4]> var_760 = const()[name = tensor<string, []>("op_760"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_15_cast = reshape(shape = var_760, x = hidden_states_107_cast)[name = tensor<string, []>("value_states_15_cast")];
            tensor<int32, [4]> var_762_perm_0 = const()[name = tensor<string, []>("op_762_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_763_transpose_x_0 = const()[name = tensor<string, []>("op_763_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_763_transpose_y_0 = const()[name = tensor<string, []>("op_763_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_91 = transpose(perm = var_762_perm_0, x = key_states_23_cast)[name = tensor<string, []>("transpose_91")];
            tensor<fp16, [1, 32, 128, 128]> var_763_cast = matmul(transpose_x = var_763_transpose_x_0, transpose_y = var_763_transpose_y_0, x = query_states_15_cast, y = transpose_91)[name = tensor<string, []>("op_763_cast")];
            tensor<fp16, []> _inversed_attn_weights_13_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_13_cast = mul(x = var_763_cast, y = _inversed_attn_weights_13_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_13_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_31_cast = add(x = _inversed_attn_weights_13_cast, y = attention_mask_cast)[name = tensor<string, []>("input_31_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_767_cast = softmax(axis = var_10, x = input_31_cast)[name = tensor<string, []>("op_767_cast")];
            tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_13_cast = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = var_767_cast, y = value_states_15_cast)[name = tensor<string, []>("attn_output_13_cast")];
            tensor<int32, [4]> var_770_perm_0 = const()[name = tensor<string, []>("op_770_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_772 = const()[name = tensor<string, []>("op_772"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_90 = transpose(perm = var_770_perm_0, x = attn_output_13_cast)[name = tensor<string, []>("transpose_90")];
            tensor<fp16, [1, 128, 2048]> input_33_cast = reshape(shape = var_772, x = transpose_90)[name = tensor<string, []>("input_33_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_3_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101511488))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103608704))), name = tensor<string, []>("model_model_layers_3_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_24_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_3_self_attn_o_proj_weight_to_fp16_palettized, x = input_33_cast)[name = tensor<string, []>("linear_24_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_111_cast = add(x = hidden_states_91_cast, y = linear_24_cast)[name = tensor<string, []>("hidden_states_111_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_7 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_7"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_779_cast = pow(x = hidden_states_111_cast, y = var_13_promoted_to_fp16_7)[name = tensor<string, []>("op_779_cast")];
            tensor<int32, [1]> var_780 = const()[name = tensor<string, []>("op_780"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_15_cast = reduce_mean(axes = var_780, keep_dims = var_23, x = var_779_cast)[name = tensor<string, []>("variance_15_cast")];
            tensor<fp16, []> var_782_to_fp16 = const()[name = tensor<string, []>("op_782_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_783_cast = add(x = variance_15_cast, y = var_782_to_fp16)[name = tensor<string, []>("op_783_cast")];
            tensor<fp16, []> var_784_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_784_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_784_cast = rsqrt(epsilon = var_784_epsilon_0_to_fp16, x = var_783_cast)[name = tensor<string, []>("op_784_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_115_cast = mul(x = hidden_states_111_cast, y = var_784_cast)[name = tensor<string, []>("hidden_states_115_cast")];
            tensor<fp16, [2048]> model_model_layers_3_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_3_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103608832)))];
            tensor<fp16, [1, 128, 2048]> input_35_cast = mul(x = model_model_layers_3_post_attention_layernorm_weight_to_fp16, y = hidden_states_115_cast)[name = tensor<string, []>("input_35_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_3_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103612992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109380224))), name = tensor<string, []>("model_model_layers_3_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_25_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_3_mlp_gate_proj_weight_to_fp16_palettized, x = input_35_cast)[name = tensor<string, []>("linear_25_cast")];
            tensor<fp16, [1, 128, 5632]> var_796_cast = silu(x = linear_25_cast)[name = tensor<string, []>("op_796_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_3_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(109380352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115147584))), name = tensor<string, []>("model_model_layers_3_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_26_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_3_mlp_up_proj_weight_to_fp16_palettized, x = input_35_cast)[name = tensor<string, []>("linear_26_cast")];
            tensor<fp16, [1, 128, 5632]> input_39_cast = mul(x = var_796_cast, y = linear_26_cast)[name = tensor<string, []>("input_39_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_3_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(115147712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120914944))), name = tensor<string, []>("model_model_layers_3_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_27_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_3_mlp_down_proj_weight_to_fp16_palettized, x = input_39_cast)[name = tensor<string, []>("linear_27_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_121_cast = add(x = hidden_states_111_cast, y = linear_27_cast)[name = tensor<string, []>("hidden_states_121_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_8 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_8"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_809_cast = pow(x = hidden_states_121_cast, y = var_13_promoted_to_fp16_8)[name = tensor<string, []>("op_809_cast")];
            tensor<int32, [1]> var_810 = const()[name = tensor<string, []>("op_810"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_17_cast = reduce_mean(axes = var_810, keep_dims = var_23, x = var_809_cast)[name = tensor<string, []>("variance_17_cast")];
            tensor<fp16, []> var_812_to_fp16 = const()[name = tensor<string, []>("op_812_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_813_cast = add(x = variance_17_cast, y = var_812_to_fp16)[name = tensor<string, []>("op_813_cast")];
            tensor<fp16, []> var_814_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_814_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_814_cast = rsqrt(epsilon = var_814_epsilon_0_to_fp16, x = var_813_cast)[name = tensor<string, []>("op_814_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_125_cast = mul(x = hidden_states_121_cast, y = var_814_cast)[name = tensor<string, []>("hidden_states_125_cast")];
            tensor<fp16, [2048]> model_model_layers_4_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_4_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120915072)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_129_cast = mul(x = model_model_layers_4_input_layernorm_weight_to_fp16, y = hidden_states_125_cast)[name = tensor<string, []>("hidden_states_129_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_4_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120919232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123016448))), name = tensor<string, []>("model_model_layers_4_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_28_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_4_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_129_cast)[name = tensor<string, []>("linear_28_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_4_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123016576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123278784))), name = tensor<string, []>("model_model_layers_4_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_29_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_4_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_129_cast)[name = tensor<string, []>("linear_29_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_4_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123278912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123541120))), name = tensor<string, []>("model_model_layers_4_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_30_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_4_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_129_cast)[name = tensor<string, []>("linear_30_cast")];
            tensor<int32, [4]> var_834 = const()[name = tensor<string, []>("op_834"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_835_cast = reshape(shape = var_834, x = linear_28_cast)[name = tensor<string, []>("op_835_cast")];
            tensor<int32, [4]> q_9_perm_0 = const()[name = tensor<string, []>("q_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_837 = const()[name = tensor<string, []>("op_837"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_838_cast = reshape(shape = var_837, x = linear_29_cast)[name = tensor<string, []>("op_838_cast")];
            tensor<int32, [4]> key_states_27_perm_0 = const()[name = tensor<string, []>("key_states_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_840 = const()[name = tensor<string, []>("op_840"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_841_cast = reshape(shape = var_840, x = linear_30_cast)[name = tensor<string, []>("op_841_cast")];
            tensor<int32, [4]> hidden_states_135_perm_0 = const()[name = tensor<string, []>("hidden_states_135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_89 = transpose(perm = q_9_perm_0, x = var_835_cast)[name = tensor<string, []>("transpose_89")];
            tensor<fp16, [1, 32, 128, 64]> var_867_cast = mul(x = transpose_89, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_867_cast")];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = tensor<string, []>("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = tensor<string, []>("x1_17_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = tensor<string, []>("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_17_cast = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = transpose_89)[name = tensor<string, []>("x1_17_cast")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = tensor<string, []>("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = tensor<string, []>("x2_17_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = tensor<string, []>("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_17_cast = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = transpose_89)[name = tensor<string, []>("x2_17_cast")];
            tensor<fp16, []> const_79_promoted_to_fp16 = const()[name = tensor<string, []>("const_79_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_878_cast = mul(x = x2_17_cast, y = const_79_promoted_to_fp16)[name = tensor<string, []>("op_878_cast")];
            tensor<bool, []> var_880_interleave_0 = const()[name = tensor<string, []>("op_880_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_880_cast = concat(axis = var_10, interleave = var_880_interleave_0, values = (var_878_cast, x1_17_cast))[name = tensor<string, []>("op_880_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_881_cast = mul(x = var_880_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_881_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_19_cast = add(x = var_867_cast, y = var_881_cast)[name = tensor<string, []>("query_states_19_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_88 = transpose(perm = key_states_27_perm_0, x = var_838_cast)[name = tensor<string, []>("transpose_88")];
            tensor<fp16, [1, 4, 128, 64]> var_883_cast = mul(x = transpose_88, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_883_cast")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = tensor<string, []>("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = tensor<string, []>("x1_19_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = tensor<string, []>("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_19_cast = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = transpose_88)[name = tensor<string, []>("x1_19_cast")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = tensor<string, []>("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = tensor<string, []>("x2_19_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = tensor<string, []>("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_19_cast = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = transpose_88)[name = tensor<string, []>("x2_19_cast")];
            tensor<fp16, []> const_82_promoted_to_fp16 = const()[name = tensor<string, []>("const_82_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_894_cast = mul(x = x2_19_cast, y = const_82_promoted_to_fp16)[name = tensor<string, []>("op_894_cast")];
            tensor<bool, []> var_896_interleave_0 = const()[name = tensor<string, []>("op_896_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_896_cast = concat(axis = var_10, interleave = var_896_interleave_0, values = (var_894_cast, x1_19_cast))[name = tensor<string, []>("op_896_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_897_cast = mul(x = var_896_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_897_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_131_cast = add(x = var_883_cast, y = var_897_cast)[name = tensor<string, []>("hidden_states_131_cast")];
            tensor<int32, [1]> var_906_axes_0 = const()[name = tensor<string, []>("op_906_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_906_cast = expand_dims(axes = var_906_axes_0, x = hidden_states_131_cast)[name = tensor<string, []>("op_906_cast")];
            tensor<int32, [5]> hidden_states_133_reps_0 = const()[name = tensor<string, []>("hidden_states_133_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_133_cast = tile(reps = hidden_states_133_reps_0, x = var_906_cast)[name = tensor<string, []>("hidden_states_133_cast")];
            tensor<int32, [4]> var_914 = const()[name = tensor<string, []>("op_914"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_29_cast = reshape(shape = var_914, x = hidden_states_133_cast)[name = tensor<string, []>("key_states_29_cast")];
            tensor<int32, [1]> var_923_axes_0 = const()[name = tensor<string, []>("op_923_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_87 = transpose(perm = hidden_states_135_perm_0, x = var_841_cast)[name = tensor<string, []>("transpose_87")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_923_cast = expand_dims(axes = var_923_axes_0, x = transpose_87)[name = tensor<string, []>("op_923_cast")];
            tensor<int32, [5]> hidden_states_137_reps_0 = const()[name = tensor<string, []>("hidden_states_137_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_137_cast = tile(reps = hidden_states_137_reps_0, x = var_923_cast)[name = tensor<string, []>("hidden_states_137_cast")];
            tensor<int32, [4]> var_931 = const()[name = tensor<string, []>("op_931"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_19_cast = reshape(shape = var_931, x = hidden_states_137_cast)[name = tensor<string, []>("value_states_19_cast")];
            tensor<int32, [4]> var_933_perm_0 = const()[name = tensor<string, []>("op_933_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_934_transpose_x_0 = const()[name = tensor<string, []>("op_934_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_934_transpose_y_0 = const()[name = tensor<string, []>("op_934_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_86 = transpose(perm = var_933_perm_0, x = key_states_29_cast)[name = tensor<string, []>("transpose_86")];
            tensor<fp16, [1, 32, 128, 128]> var_934_cast = matmul(transpose_x = var_934_transpose_x_0, transpose_y = var_934_transpose_y_0, x = query_states_19_cast, y = transpose_86)[name = tensor<string, []>("op_934_cast")];
            tensor<fp16, []> _inversed_attn_weights_17_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_17_cast = mul(x = var_934_cast, y = _inversed_attn_weights_17_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_17_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_41_cast = add(x = _inversed_attn_weights_17_cast, y = attention_mask_cast)[name = tensor<string, []>("input_41_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_938_cast = softmax(axis = var_10, x = input_41_cast)[name = tensor<string, []>("op_938_cast")];
            tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_17_cast = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = var_938_cast, y = value_states_19_cast)[name = tensor<string, []>("attn_output_17_cast")];
            tensor<int32, [4]> var_941_perm_0 = const()[name = tensor<string, []>("op_941_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_943 = const()[name = tensor<string, []>("op_943"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_85 = transpose(perm = var_941_perm_0, x = attn_output_17_cast)[name = tensor<string, []>("transpose_85")];
            tensor<fp16, [1, 128, 2048]> input_43_cast = reshape(shape = var_943, x = transpose_85)[name = tensor<string, []>("input_43_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_4_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123541248))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125638464))), name = tensor<string, []>("model_model_layers_4_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_31_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_4_self_attn_o_proj_weight_to_fp16_palettized, x = input_43_cast)[name = tensor<string, []>("linear_31_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_141_cast = add(x = hidden_states_121_cast, y = linear_31_cast)[name = tensor<string, []>("hidden_states_141_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_9 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_9"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_950_cast = pow(x = hidden_states_141_cast, y = var_13_promoted_to_fp16_9)[name = tensor<string, []>("op_950_cast")];
            tensor<int32, [1]> var_951 = const()[name = tensor<string, []>("op_951"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_19_cast = reduce_mean(axes = var_951, keep_dims = var_23, x = var_950_cast)[name = tensor<string, []>("variance_19_cast")];
            tensor<fp16, []> var_953_to_fp16 = const()[name = tensor<string, []>("op_953_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_954_cast = add(x = variance_19_cast, y = var_953_to_fp16)[name = tensor<string, []>("op_954_cast")];
            tensor<fp16, []> var_955_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_955_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_955_cast = rsqrt(epsilon = var_955_epsilon_0_to_fp16, x = var_954_cast)[name = tensor<string, []>("op_955_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_145_cast = mul(x = hidden_states_141_cast, y = var_955_cast)[name = tensor<string, []>("hidden_states_145_cast")];
            tensor<fp16, [2048]> model_model_layers_4_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_4_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125638592)))];
            tensor<fp16, [1, 128, 2048]> input_45_cast = mul(x = model_model_layers_4_post_attention_layernorm_weight_to_fp16, y = hidden_states_145_cast)[name = tensor<string, []>("input_45_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_4_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125642752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131409984))), name = tensor<string, []>("model_model_layers_4_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_32_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_4_mlp_gate_proj_weight_to_fp16_palettized, x = input_45_cast)[name = tensor<string, []>("linear_32_cast")];
            tensor<fp16, [1, 128, 5632]> var_967_cast = silu(x = linear_32_cast)[name = tensor<string, []>("op_967_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_4_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131410112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137177344))), name = tensor<string, []>("model_model_layers_4_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_33_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_4_mlp_up_proj_weight_to_fp16_palettized, x = input_45_cast)[name = tensor<string, []>("linear_33_cast")];
            tensor<fp16, [1, 128, 5632]> input_49_cast = mul(x = var_967_cast, y = linear_33_cast)[name = tensor<string, []>("input_49_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_4_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137177472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142944704))), name = tensor<string, []>("model_model_layers_4_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_34_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_4_mlp_down_proj_weight_to_fp16_palettized, x = input_49_cast)[name = tensor<string, []>("linear_34_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_151_cast = add(x = hidden_states_141_cast, y = linear_34_cast)[name = tensor<string, []>("hidden_states_151_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_10 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_10"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_980_cast = pow(x = hidden_states_151_cast, y = var_13_promoted_to_fp16_10)[name = tensor<string, []>("op_980_cast")];
            tensor<int32, [1]> var_981 = const()[name = tensor<string, []>("op_981"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_21_cast = reduce_mean(axes = var_981, keep_dims = var_23, x = var_980_cast)[name = tensor<string, []>("variance_21_cast")];
            tensor<fp16, []> var_983_to_fp16 = const()[name = tensor<string, []>("op_983_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_984_cast = add(x = variance_21_cast, y = var_983_to_fp16)[name = tensor<string, []>("op_984_cast")];
            tensor<fp16, []> var_985_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_985_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_985_cast = rsqrt(epsilon = var_985_epsilon_0_to_fp16, x = var_984_cast)[name = tensor<string, []>("op_985_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_155_cast = mul(x = hidden_states_151_cast, y = var_985_cast)[name = tensor<string, []>("hidden_states_155_cast")];
            tensor<fp16, [2048]> model_model_layers_5_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_5_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142944832)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_159_cast = mul(x = model_model_layers_5_input_layernorm_weight_to_fp16, y = hidden_states_155_cast)[name = tensor<string, []>("hidden_states_159_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_5_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142948992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145046208))), name = tensor<string, []>("model_model_layers_5_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_35_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_5_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_159_cast)[name = tensor<string, []>("linear_35_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_5_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145046336))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145308544))), name = tensor<string, []>("model_model_layers_5_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_36_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_5_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_159_cast)[name = tensor<string, []>("linear_36_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_5_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145308672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145570880))), name = tensor<string, []>("model_model_layers_5_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_37_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_5_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_159_cast)[name = tensor<string, []>("linear_37_cast")];
            tensor<int32, [4]> var_1005 = const()[name = tensor<string, []>("op_1005"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1006_cast = reshape(shape = var_1005, x = linear_35_cast)[name = tensor<string, []>("op_1006_cast")];
            tensor<int32, [4]> q_11_perm_0 = const()[name = tensor<string, []>("q_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1008 = const()[name = tensor<string, []>("op_1008"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1009_cast = reshape(shape = var_1008, x = linear_36_cast)[name = tensor<string, []>("op_1009_cast")];
            tensor<int32, [4]> key_states_33_perm_0 = const()[name = tensor<string, []>("key_states_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1011 = const()[name = tensor<string, []>("op_1011"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1012_cast = reshape(shape = var_1011, x = linear_37_cast)[name = tensor<string, []>("op_1012_cast")];
            tensor<int32, [4]> hidden_states_165_perm_0 = const()[name = tensor<string, []>("hidden_states_165_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_84 = transpose(perm = q_11_perm_0, x = var_1006_cast)[name = tensor<string, []>("transpose_84")];
            tensor<fp16, [1, 32, 128, 64]> var_1038_cast = mul(x = transpose_84, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1038_cast")];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = tensor<string, []>("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = tensor<string, []>("x1_21_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = tensor<string, []>("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_21_cast = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = transpose_84)[name = tensor<string, []>("x1_21_cast")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = tensor<string, []>("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = tensor<string, []>("x2_21_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = tensor<string, []>("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_21_cast = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = transpose_84)[name = tensor<string, []>("x2_21_cast")];
            tensor<fp16, []> const_96_promoted_to_fp16 = const()[name = tensor<string, []>("const_96_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1049_cast = mul(x = x2_21_cast, y = const_96_promoted_to_fp16)[name = tensor<string, []>("op_1049_cast")];
            tensor<bool, []> var_1051_interleave_0 = const()[name = tensor<string, []>("op_1051_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1051_cast = concat(axis = var_10, interleave = var_1051_interleave_0, values = (var_1049_cast, x1_21_cast))[name = tensor<string, []>("op_1051_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1052_cast = mul(x = var_1051_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1052_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_23_cast = add(x = var_1038_cast, y = var_1052_cast)[name = tensor<string, []>("query_states_23_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_83 = transpose(perm = key_states_33_perm_0, x = var_1009_cast)[name = tensor<string, []>("transpose_83")];
            tensor<fp16, [1, 4, 128, 64]> var_1054_cast = mul(x = transpose_83, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1054_cast")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = tensor<string, []>("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = tensor<string, []>("x1_23_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = tensor<string, []>("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_23_cast = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = transpose_83)[name = tensor<string, []>("x1_23_cast")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = tensor<string, []>("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = tensor<string, []>("x2_23_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = tensor<string, []>("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_23_cast = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = transpose_83)[name = tensor<string, []>("x2_23_cast")];
            tensor<fp16, []> const_99_promoted_to_fp16 = const()[name = tensor<string, []>("const_99_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1065_cast = mul(x = x2_23_cast, y = const_99_promoted_to_fp16)[name = tensor<string, []>("op_1065_cast")];
            tensor<bool, []> var_1067_interleave_0 = const()[name = tensor<string, []>("op_1067_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1067_cast = concat(axis = var_10, interleave = var_1067_interleave_0, values = (var_1065_cast, x1_23_cast))[name = tensor<string, []>("op_1067_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1068_cast = mul(x = var_1067_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1068_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_161_cast = add(x = var_1054_cast, y = var_1068_cast)[name = tensor<string, []>("hidden_states_161_cast")];
            tensor<int32, [1]> var_1077_axes_0 = const()[name = tensor<string, []>("op_1077_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1077_cast = expand_dims(axes = var_1077_axes_0, x = hidden_states_161_cast)[name = tensor<string, []>("op_1077_cast")];
            tensor<int32, [5]> hidden_states_163_reps_0 = const()[name = tensor<string, []>("hidden_states_163_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_163_cast = tile(reps = hidden_states_163_reps_0, x = var_1077_cast)[name = tensor<string, []>("hidden_states_163_cast")];
            tensor<int32, [4]> var_1085 = const()[name = tensor<string, []>("op_1085"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_35_cast = reshape(shape = var_1085, x = hidden_states_163_cast)[name = tensor<string, []>("key_states_35_cast")];
            tensor<int32, [1]> var_1094_axes_0 = const()[name = tensor<string, []>("op_1094_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_82 = transpose(perm = hidden_states_165_perm_0, x = var_1012_cast)[name = tensor<string, []>("transpose_82")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1094_cast = expand_dims(axes = var_1094_axes_0, x = transpose_82)[name = tensor<string, []>("op_1094_cast")];
            tensor<int32, [5]> hidden_states_167_reps_0 = const()[name = tensor<string, []>("hidden_states_167_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_167_cast = tile(reps = hidden_states_167_reps_0, x = var_1094_cast)[name = tensor<string, []>("hidden_states_167_cast")];
            tensor<int32, [4]> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_23_cast = reshape(shape = var_1102, x = hidden_states_167_cast)[name = tensor<string, []>("value_states_23_cast")];
            tensor<int32, [4]> var_1104_perm_0 = const()[name = tensor<string, []>("op_1104_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1105_transpose_x_0 = const()[name = tensor<string, []>("op_1105_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1105_transpose_y_0 = const()[name = tensor<string, []>("op_1105_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_81 = transpose(perm = var_1104_perm_0, x = key_states_35_cast)[name = tensor<string, []>("transpose_81")];
            tensor<fp16, [1, 32, 128, 128]> var_1105_cast = matmul(transpose_x = var_1105_transpose_x_0, transpose_y = var_1105_transpose_y_0, x = query_states_23_cast, y = transpose_81)[name = tensor<string, []>("op_1105_cast")];
            tensor<fp16, []> _inversed_attn_weights_21_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_21_cast = mul(x = var_1105_cast, y = _inversed_attn_weights_21_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_21_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_51_cast = add(x = _inversed_attn_weights_21_cast, y = attention_mask_cast)[name = tensor<string, []>("input_51_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1109_cast = softmax(axis = var_10, x = input_51_cast)[name = tensor<string, []>("op_1109_cast")];
            tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_21_cast = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = var_1109_cast, y = value_states_23_cast)[name = tensor<string, []>("attn_output_21_cast")];
            tensor<int32, [4]> var_1112_perm_0 = const()[name = tensor<string, []>("op_1112_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1114 = const()[name = tensor<string, []>("op_1114"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_80 = transpose(perm = var_1112_perm_0, x = attn_output_21_cast)[name = tensor<string, []>("transpose_80")];
            tensor<fp16, [1, 128, 2048]> input_53_cast = reshape(shape = var_1114, x = transpose_80)[name = tensor<string, []>("input_53_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_5_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145571008))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147668224))), name = tensor<string, []>("model_model_layers_5_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_38_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_5_self_attn_o_proj_weight_to_fp16_palettized, x = input_53_cast)[name = tensor<string, []>("linear_38_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_171_cast = add(x = hidden_states_151_cast, y = linear_38_cast)[name = tensor<string, []>("hidden_states_171_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_11 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_11"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1121_cast = pow(x = hidden_states_171_cast, y = var_13_promoted_to_fp16_11)[name = tensor<string, []>("op_1121_cast")];
            tensor<int32, [1]> var_1122 = const()[name = tensor<string, []>("op_1122"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_23_cast = reduce_mean(axes = var_1122, keep_dims = var_23, x = var_1121_cast)[name = tensor<string, []>("variance_23_cast")];
            tensor<fp16, []> var_1124_to_fp16 = const()[name = tensor<string, []>("op_1124_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1125_cast = add(x = variance_23_cast, y = var_1124_to_fp16)[name = tensor<string, []>("op_1125_cast")];
            tensor<fp16, []> var_1126_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1126_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1126_cast = rsqrt(epsilon = var_1126_epsilon_0_to_fp16, x = var_1125_cast)[name = tensor<string, []>("op_1126_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_175_cast = mul(x = hidden_states_171_cast, y = var_1126_cast)[name = tensor<string, []>("hidden_states_175_cast")];
            tensor<fp16, [2048]> model_model_layers_5_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_5_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147668352)))];
            tensor<fp16, [1, 128, 2048]> input_55_cast = mul(x = model_model_layers_5_post_attention_layernorm_weight_to_fp16, y = hidden_states_175_cast)[name = tensor<string, []>("input_55_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_5_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147672512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153439744))), name = tensor<string, []>("model_model_layers_5_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_39_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_5_mlp_gate_proj_weight_to_fp16_palettized, x = input_55_cast)[name = tensor<string, []>("linear_39_cast")];
            tensor<fp16, [1, 128, 5632]> var_1138_cast = silu(x = linear_39_cast)[name = tensor<string, []>("op_1138_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_5_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153439872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159207104))), name = tensor<string, []>("model_model_layers_5_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_40_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_5_mlp_up_proj_weight_to_fp16_palettized, x = input_55_cast)[name = tensor<string, []>("linear_40_cast")];
            tensor<fp16, [1, 128, 5632]> input_59_cast = mul(x = var_1138_cast, y = linear_40_cast)[name = tensor<string, []>("input_59_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_5_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159207232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164974464))), name = tensor<string, []>("model_model_layers_5_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_41_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_5_mlp_down_proj_weight_to_fp16_palettized, x = input_59_cast)[name = tensor<string, []>("linear_41_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_181_cast = add(x = hidden_states_171_cast, y = linear_41_cast)[name = tensor<string, []>("hidden_states_181_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_12 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_12"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1151_cast = pow(x = hidden_states_181_cast, y = var_13_promoted_to_fp16_12)[name = tensor<string, []>("op_1151_cast")];
            tensor<int32, [1]> var_1152 = const()[name = tensor<string, []>("op_1152"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_25_cast = reduce_mean(axes = var_1152, keep_dims = var_23, x = var_1151_cast)[name = tensor<string, []>("variance_25_cast")];
            tensor<fp16, []> var_1154_to_fp16 = const()[name = tensor<string, []>("op_1154_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1155_cast = add(x = variance_25_cast, y = var_1154_to_fp16)[name = tensor<string, []>("op_1155_cast")];
            tensor<fp16, []> var_1156_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1156_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1156_cast = rsqrt(epsilon = var_1156_epsilon_0_to_fp16, x = var_1155_cast)[name = tensor<string, []>("op_1156_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_185_cast = mul(x = hidden_states_181_cast, y = var_1156_cast)[name = tensor<string, []>("hidden_states_185_cast")];
            tensor<fp16, [2048]> model_model_layers_6_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_6_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164974592)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_189_cast = mul(x = model_model_layers_6_input_layernorm_weight_to_fp16, y = hidden_states_185_cast)[name = tensor<string, []>("hidden_states_189_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_6_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164978752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167075968))), name = tensor<string, []>("model_model_layers_6_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_42_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_6_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_189_cast)[name = tensor<string, []>("linear_42_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_6_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167076096))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167338304))), name = tensor<string, []>("model_model_layers_6_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_43_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_6_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_189_cast)[name = tensor<string, []>("linear_43_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_6_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167338432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167600640))), name = tensor<string, []>("model_model_layers_6_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_44_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_6_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_189_cast)[name = tensor<string, []>("linear_44_cast")];
            tensor<int32, [4]> var_1176 = const()[name = tensor<string, []>("op_1176"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1177_cast = reshape(shape = var_1176, x = linear_42_cast)[name = tensor<string, []>("op_1177_cast")];
            tensor<int32, [4]> q_13_perm_0 = const()[name = tensor<string, []>("q_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1179 = const()[name = tensor<string, []>("op_1179"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1180_cast = reshape(shape = var_1179, x = linear_43_cast)[name = tensor<string, []>("op_1180_cast")];
            tensor<int32, [4]> key_states_39_perm_0 = const()[name = tensor<string, []>("key_states_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1182 = const()[name = tensor<string, []>("op_1182"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1183_cast = reshape(shape = var_1182, x = linear_44_cast)[name = tensor<string, []>("op_1183_cast")];
            tensor<int32, [4]> hidden_states_195_perm_0 = const()[name = tensor<string, []>("hidden_states_195_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_79 = transpose(perm = q_13_perm_0, x = var_1177_cast)[name = tensor<string, []>("transpose_79")];
            tensor<fp16, [1, 32, 128, 64]> var_1209_cast = mul(x = transpose_79, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1209_cast")];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = tensor<string, []>("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = tensor<string, []>("x1_25_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = tensor<string, []>("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_25_cast = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = transpose_79)[name = tensor<string, []>("x1_25_cast")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = tensor<string, []>("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = tensor<string, []>("x2_25_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = tensor<string, []>("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_25_cast = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = transpose_79)[name = tensor<string, []>("x2_25_cast")];
            tensor<fp16, []> const_113_promoted_to_fp16 = const()[name = tensor<string, []>("const_113_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1220_cast = mul(x = x2_25_cast, y = const_113_promoted_to_fp16)[name = tensor<string, []>("op_1220_cast")];
            tensor<bool, []> var_1222_interleave_0 = const()[name = tensor<string, []>("op_1222_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1222_cast = concat(axis = var_10, interleave = var_1222_interleave_0, values = (var_1220_cast, x1_25_cast))[name = tensor<string, []>("op_1222_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1223_cast = mul(x = var_1222_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1223_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_27_cast = add(x = var_1209_cast, y = var_1223_cast)[name = tensor<string, []>("query_states_27_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_78 = transpose(perm = key_states_39_perm_0, x = var_1180_cast)[name = tensor<string, []>("transpose_78")];
            tensor<fp16, [1, 4, 128, 64]> var_1225_cast = mul(x = transpose_78, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1225_cast")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = tensor<string, []>("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = tensor<string, []>("x1_27_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = tensor<string, []>("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_27_cast = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = transpose_78)[name = tensor<string, []>("x1_27_cast")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = tensor<string, []>("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = tensor<string, []>("x2_27_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = tensor<string, []>("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_27_cast = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = transpose_78)[name = tensor<string, []>("x2_27_cast")];
            tensor<fp16, []> const_116_promoted_to_fp16 = const()[name = tensor<string, []>("const_116_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1236_cast = mul(x = x2_27_cast, y = const_116_promoted_to_fp16)[name = tensor<string, []>("op_1236_cast")];
            tensor<bool, []> var_1238_interleave_0 = const()[name = tensor<string, []>("op_1238_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1238_cast = concat(axis = var_10, interleave = var_1238_interleave_0, values = (var_1236_cast, x1_27_cast))[name = tensor<string, []>("op_1238_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1239_cast = mul(x = var_1238_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1239_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_191_cast = add(x = var_1225_cast, y = var_1239_cast)[name = tensor<string, []>("hidden_states_191_cast")];
            tensor<int32, [1]> var_1248_axes_0 = const()[name = tensor<string, []>("op_1248_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1248_cast = expand_dims(axes = var_1248_axes_0, x = hidden_states_191_cast)[name = tensor<string, []>("op_1248_cast")];
            tensor<int32, [5]> hidden_states_193_reps_0 = const()[name = tensor<string, []>("hidden_states_193_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_193_cast = tile(reps = hidden_states_193_reps_0, x = var_1248_cast)[name = tensor<string, []>("hidden_states_193_cast")];
            tensor<int32, [4]> var_1256 = const()[name = tensor<string, []>("op_1256"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_41_cast = reshape(shape = var_1256, x = hidden_states_193_cast)[name = tensor<string, []>("key_states_41_cast")];
            tensor<int32, [1]> var_1265_axes_0 = const()[name = tensor<string, []>("op_1265_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_77 = transpose(perm = hidden_states_195_perm_0, x = var_1183_cast)[name = tensor<string, []>("transpose_77")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1265_cast = expand_dims(axes = var_1265_axes_0, x = transpose_77)[name = tensor<string, []>("op_1265_cast")];
            tensor<int32, [5]> hidden_states_197_reps_0 = const()[name = tensor<string, []>("hidden_states_197_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_197_cast = tile(reps = hidden_states_197_reps_0, x = var_1265_cast)[name = tensor<string, []>("hidden_states_197_cast")];
            tensor<int32, [4]> var_1273 = const()[name = tensor<string, []>("op_1273"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_27_cast = reshape(shape = var_1273, x = hidden_states_197_cast)[name = tensor<string, []>("value_states_27_cast")];
            tensor<int32, [4]> var_1275_perm_0 = const()[name = tensor<string, []>("op_1275_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1276_transpose_x_0 = const()[name = tensor<string, []>("op_1276_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1276_transpose_y_0 = const()[name = tensor<string, []>("op_1276_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_76 = transpose(perm = var_1275_perm_0, x = key_states_41_cast)[name = tensor<string, []>("transpose_76")];
            tensor<fp16, [1, 32, 128, 128]> var_1276_cast = matmul(transpose_x = var_1276_transpose_x_0, transpose_y = var_1276_transpose_y_0, x = query_states_27_cast, y = transpose_76)[name = tensor<string, []>("op_1276_cast")];
            tensor<fp16, []> _inversed_attn_weights_25_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_25_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_25_cast = mul(x = var_1276_cast, y = _inversed_attn_weights_25_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_25_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_61_cast = add(x = _inversed_attn_weights_25_cast, y = attention_mask_cast)[name = tensor<string, []>("input_61_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1280_cast = softmax(axis = var_10, x = input_61_cast)[name = tensor<string, []>("op_1280_cast")];
            tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_25_cast = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = var_1280_cast, y = value_states_27_cast)[name = tensor<string, []>("attn_output_25_cast")];
            tensor<int32, [4]> var_1283_perm_0 = const()[name = tensor<string, []>("op_1283_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1285 = const()[name = tensor<string, []>("op_1285"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_75 = transpose(perm = var_1283_perm_0, x = attn_output_25_cast)[name = tensor<string, []>("transpose_75")];
            tensor<fp16, [1, 128, 2048]> input_63_cast = reshape(shape = var_1285, x = transpose_75)[name = tensor<string, []>("input_63_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_6_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167600768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169697984))), name = tensor<string, []>("model_model_layers_6_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_45_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_6_self_attn_o_proj_weight_to_fp16_palettized, x = input_63_cast)[name = tensor<string, []>("linear_45_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_201_cast = add(x = hidden_states_181_cast, y = linear_45_cast)[name = tensor<string, []>("hidden_states_201_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_13 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_13"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1292_cast = pow(x = hidden_states_201_cast, y = var_13_promoted_to_fp16_13)[name = tensor<string, []>("op_1292_cast")];
            tensor<int32, [1]> var_1293 = const()[name = tensor<string, []>("op_1293"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_27_cast = reduce_mean(axes = var_1293, keep_dims = var_23, x = var_1292_cast)[name = tensor<string, []>("variance_27_cast")];
            tensor<fp16, []> var_1295_to_fp16 = const()[name = tensor<string, []>("op_1295_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1296_cast = add(x = variance_27_cast, y = var_1295_to_fp16)[name = tensor<string, []>("op_1296_cast")];
            tensor<fp16, []> var_1297_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1297_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1297_cast = rsqrt(epsilon = var_1297_epsilon_0_to_fp16, x = var_1296_cast)[name = tensor<string, []>("op_1297_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_205_cast = mul(x = hidden_states_201_cast, y = var_1297_cast)[name = tensor<string, []>("hidden_states_205_cast")];
            tensor<fp16, [2048]> model_model_layers_6_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_6_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169698112)))];
            tensor<fp16, [1, 128, 2048]> input_65_cast = mul(x = model_model_layers_6_post_attention_layernorm_weight_to_fp16, y = hidden_states_205_cast)[name = tensor<string, []>("input_65_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_6_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169702272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175469504))), name = tensor<string, []>("model_model_layers_6_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_46_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_6_mlp_gate_proj_weight_to_fp16_palettized, x = input_65_cast)[name = tensor<string, []>("linear_46_cast")];
            tensor<fp16, [1, 128, 5632]> var_1309_cast = silu(x = linear_46_cast)[name = tensor<string, []>("op_1309_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_6_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175469632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181236864))), name = tensor<string, []>("model_model_layers_6_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_47_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_6_mlp_up_proj_weight_to_fp16_palettized, x = input_65_cast)[name = tensor<string, []>("linear_47_cast")];
            tensor<fp16, [1, 128, 5632]> input_69_cast = mul(x = var_1309_cast, y = linear_47_cast)[name = tensor<string, []>("input_69_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_6_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181236992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187004224))), name = tensor<string, []>("model_model_layers_6_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_48_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_6_mlp_down_proj_weight_to_fp16_palettized, x = input_69_cast)[name = tensor<string, []>("linear_48_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_211_cast = add(x = hidden_states_201_cast, y = linear_48_cast)[name = tensor<string, []>("hidden_states_211_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_14 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_14"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1322_cast = pow(x = hidden_states_211_cast, y = var_13_promoted_to_fp16_14)[name = tensor<string, []>("op_1322_cast")];
            tensor<int32, [1]> var_1323 = const()[name = tensor<string, []>("op_1323"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_29_cast = reduce_mean(axes = var_1323, keep_dims = var_23, x = var_1322_cast)[name = tensor<string, []>("variance_29_cast")];
            tensor<fp16, []> var_1325_to_fp16 = const()[name = tensor<string, []>("op_1325_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1326_cast = add(x = variance_29_cast, y = var_1325_to_fp16)[name = tensor<string, []>("op_1326_cast")];
            tensor<fp16, []> var_1327_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1327_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1327_cast = rsqrt(epsilon = var_1327_epsilon_0_to_fp16, x = var_1326_cast)[name = tensor<string, []>("op_1327_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_215_cast = mul(x = hidden_states_211_cast, y = var_1327_cast)[name = tensor<string, []>("hidden_states_215_cast")];
            tensor<fp16, [2048]> model_model_layers_7_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_7_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187004352)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_219_cast = mul(x = model_model_layers_7_input_layernorm_weight_to_fp16, y = hidden_states_215_cast)[name = tensor<string, []>("hidden_states_219_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_7_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187008512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189105728))), name = tensor<string, []>("model_model_layers_7_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_49_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_7_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_219_cast)[name = tensor<string, []>("linear_49_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_7_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189105856))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189368064))), name = tensor<string, []>("model_model_layers_7_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_50_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_7_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_219_cast)[name = tensor<string, []>("linear_50_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_7_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189368192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189630400))), name = tensor<string, []>("model_model_layers_7_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_51_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_7_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_219_cast)[name = tensor<string, []>("linear_51_cast")];
            tensor<int32, [4]> var_1347 = const()[name = tensor<string, []>("op_1347"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1348_cast = reshape(shape = var_1347, x = linear_49_cast)[name = tensor<string, []>("op_1348_cast")];
            tensor<int32, [4]> q_15_perm_0 = const()[name = tensor<string, []>("q_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1350 = const()[name = tensor<string, []>("op_1350"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1351_cast = reshape(shape = var_1350, x = linear_50_cast)[name = tensor<string, []>("op_1351_cast")];
            tensor<int32, [4]> key_states_45_perm_0 = const()[name = tensor<string, []>("key_states_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1353 = const()[name = tensor<string, []>("op_1353"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1354_cast = reshape(shape = var_1353, x = linear_51_cast)[name = tensor<string, []>("op_1354_cast")];
            tensor<int32, [4]> hidden_states_225_perm_0 = const()[name = tensor<string, []>("hidden_states_225_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_74 = transpose(perm = q_15_perm_0, x = var_1348_cast)[name = tensor<string, []>("transpose_74")];
            tensor<fp16, [1, 32, 128, 64]> var_1380_cast = mul(x = transpose_74, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1380_cast")];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = tensor<string, []>("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = tensor<string, []>("x1_29_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = tensor<string, []>("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_29_cast = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = transpose_74)[name = tensor<string, []>("x1_29_cast")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = tensor<string, []>("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = tensor<string, []>("x2_29_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = tensor<string, []>("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_29_cast = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = transpose_74)[name = tensor<string, []>("x2_29_cast")];
            tensor<fp16, []> const_130_promoted_to_fp16 = const()[name = tensor<string, []>("const_130_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1391_cast = mul(x = x2_29_cast, y = const_130_promoted_to_fp16)[name = tensor<string, []>("op_1391_cast")];
            tensor<bool, []> var_1393_interleave_0 = const()[name = tensor<string, []>("op_1393_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1393_cast = concat(axis = var_10, interleave = var_1393_interleave_0, values = (var_1391_cast, x1_29_cast))[name = tensor<string, []>("op_1393_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1394_cast = mul(x = var_1393_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1394_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_31_cast = add(x = var_1380_cast, y = var_1394_cast)[name = tensor<string, []>("query_states_31_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_73 = transpose(perm = key_states_45_perm_0, x = var_1351_cast)[name = tensor<string, []>("transpose_73")];
            tensor<fp16, [1, 4, 128, 64]> var_1396_cast = mul(x = transpose_73, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1396_cast")];
            tensor<int32, [4]> x1_31_begin_0 = const()[name = tensor<string, []>("x1_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_31_end_0 = const()[name = tensor<string, []>("x1_31_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_31_end_mask_0 = const()[name = tensor<string, []>("x1_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_31_cast = slice_by_index(begin = x1_31_begin_0, end = x1_31_end_0, end_mask = x1_31_end_mask_0, x = transpose_73)[name = tensor<string, []>("x1_31_cast")];
            tensor<int32, [4]> x2_31_begin_0 = const()[name = tensor<string, []>("x2_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_31_end_0 = const()[name = tensor<string, []>("x2_31_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_31_end_mask_0 = const()[name = tensor<string, []>("x2_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_31_cast = slice_by_index(begin = x2_31_begin_0, end = x2_31_end_0, end_mask = x2_31_end_mask_0, x = transpose_73)[name = tensor<string, []>("x2_31_cast")];
            tensor<fp16, []> const_133_promoted_to_fp16 = const()[name = tensor<string, []>("const_133_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1407_cast = mul(x = x2_31_cast, y = const_133_promoted_to_fp16)[name = tensor<string, []>("op_1407_cast")];
            tensor<bool, []> var_1409_interleave_0 = const()[name = tensor<string, []>("op_1409_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1409_cast = concat(axis = var_10, interleave = var_1409_interleave_0, values = (var_1407_cast, x1_31_cast))[name = tensor<string, []>("op_1409_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1410_cast = mul(x = var_1409_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1410_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_221_cast = add(x = var_1396_cast, y = var_1410_cast)[name = tensor<string, []>("hidden_states_221_cast")];
            tensor<int32, [1]> var_1419_axes_0 = const()[name = tensor<string, []>("op_1419_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1419_cast = expand_dims(axes = var_1419_axes_0, x = hidden_states_221_cast)[name = tensor<string, []>("op_1419_cast")];
            tensor<int32, [5]> hidden_states_223_reps_0 = const()[name = tensor<string, []>("hidden_states_223_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_223_cast = tile(reps = hidden_states_223_reps_0, x = var_1419_cast)[name = tensor<string, []>("hidden_states_223_cast")];
            tensor<int32, [4]> var_1427 = const()[name = tensor<string, []>("op_1427"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_47_cast = reshape(shape = var_1427, x = hidden_states_223_cast)[name = tensor<string, []>("key_states_47_cast")];
            tensor<int32, [1]> var_1436_axes_0 = const()[name = tensor<string, []>("op_1436_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_72 = transpose(perm = hidden_states_225_perm_0, x = var_1354_cast)[name = tensor<string, []>("transpose_72")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1436_cast = expand_dims(axes = var_1436_axes_0, x = transpose_72)[name = tensor<string, []>("op_1436_cast")];
            tensor<int32, [5]> hidden_states_227_reps_0 = const()[name = tensor<string, []>("hidden_states_227_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_227_cast = tile(reps = hidden_states_227_reps_0, x = var_1436_cast)[name = tensor<string, []>("hidden_states_227_cast")];
            tensor<int32, [4]> var_1444 = const()[name = tensor<string, []>("op_1444"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_31_cast = reshape(shape = var_1444, x = hidden_states_227_cast)[name = tensor<string, []>("value_states_31_cast")];
            tensor<int32, [4]> var_1446_perm_0 = const()[name = tensor<string, []>("op_1446_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1447_transpose_x_0 = const()[name = tensor<string, []>("op_1447_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1447_transpose_y_0 = const()[name = tensor<string, []>("op_1447_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_71 = transpose(perm = var_1446_perm_0, x = key_states_47_cast)[name = tensor<string, []>("transpose_71")];
            tensor<fp16, [1, 32, 128, 128]> var_1447_cast = matmul(transpose_x = var_1447_transpose_x_0, transpose_y = var_1447_transpose_y_0, x = query_states_31_cast, y = transpose_71)[name = tensor<string, []>("op_1447_cast")];
            tensor<fp16, []> _inversed_attn_weights_29_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_29_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_29_cast = mul(x = var_1447_cast, y = _inversed_attn_weights_29_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_29_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_71_cast = add(x = _inversed_attn_weights_29_cast, y = attention_mask_cast)[name = tensor<string, []>("input_71_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1451_cast = softmax(axis = var_10, x = input_71_cast)[name = tensor<string, []>("op_1451_cast")];
            tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_29_cast = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = var_1451_cast, y = value_states_31_cast)[name = tensor<string, []>("attn_output_29_cast")];
            tensor<int32, [4]> var_1454_perm_0 = const()[name = tensor<string, []>("op_1454_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1456 = const()[name = tensor<string, []>("op_1456"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_70 = transpose(perm = var_1454_perm_0, x = attn_output_29_cast)[name = tensor<string, []>("transpose_70")];
            tensor<fp16, [1, 128, 2048]> input_73_cast = reshape(shape = var_1456, x = transpose_70)[name = tensor<string, []>("input_73_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_7_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189630528))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191727744))), name = tensor<string, []>("model_model_layers_7_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_52_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_7_self_attn_o_proj_weight_to_fp16_palettized, x = input_73_cast)[name = tensor<string, []>("linear_52_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_231_cast = add(x = hidden_states_211_cast, y = linear_52_cast)[name = tensor<string, []>("hidden_states_231_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_15 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_15"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1463_cast = pow(x = hidden_states_231_cast, y = var_13_promoted_to_fp16_15)[name = tensor<string, []>("op_1463_cast")];
            tensor<int32, [1]> var_1464 = const()[name = tensor<string, []>("op_1464"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_31_cast = reduce_mean(axes = var_1464, keep_dims = var_23, x = var_1463_cast)[name = tensor<string, []>("variance_31_cast")];
            tensor<fp16, []> var_1466_to_fp16 = const()[name = tensor<string, []>("op_1466_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1467_cast = add(x = variance_31_cast, y = var_1466_to_fp16)[name = tensor<string, []>("op_1467_cast")];
            tensor<fp16, []> var_1468_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1468_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1468_cast = rsqrt(epsilon = var_1468_epsilon_0_to_fp16, x = var_1467_cast)[name = tensor<string, []>("op_1468_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_235_cast = mul(x = hidden_states_231_cast, y = var_1468_cast)[name = tensor<string, []>("hidden_states_235_cast")];
            tensor<fp16, [2048]> model_model_layers_7_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_7_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191727872)))];
            tensor<fp16, [1, 128, 2048]> input_75_cast = mul(x = model_model_layers_7_post_attention_layernorm_weight_to_fp16, y = hidden_states_235_cast)[name = tensor<string, []>("input_75_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_7_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191732032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197499264))), name = tensor<string, []>("model_model_layers_7_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_53_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_7_mlp_gate_proj_weight_to_fp16_palettized, x = input_75_cast)[name = tensor<string, []>("linear_53_cast")];
            tensor<fp16, [1, 128, 5632]> var_1480_cast = silu(x = linear_53_cast)[name = tensor<string, []>("op_1480_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_7_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197499392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203266624))), name = tensor<string, []>("model_model_layers_7_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_54_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_7_mlp_up_proj_weight_to_fp16_palettized, x = input_75_cast)[name = tensor<string, []>("linear_54_cast")];
            tensor<fp16, [1, 128, 5632]> input_79_cast = mul(x = var_1480_cast, y = linear_54_cast)[name = tensor<string, []>("input_79_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_7_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203266752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209033984))), name = tensor<string, []>("model_model_layers_7_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_55_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_7_mlp_down_proj_weight_to_fp16_palettized, x = input_79_cast)[name = tensor<string, []>("linear_55_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_241_cast = add(x = hidden_states_231_cast, y = linear_55_cast)[name = tensor<string, []>("hidden_states_241_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_16 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1493_cast = pow(x = hidden_states_241_cast, y = var_13_promoted_to_fp16_16)[name = tensor<string, []>("op_1493_cast")];
            tensor<int32, [1]> var_1494 = const()[name = tensor<string, []>("op_1494"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_33_cast = reduce_mean(axes = var_1494, keep_dims = var_23, x = var_1493_cast)[name = tensor<string, []>("variance_33_cast")];
            tensor<fp16, []> var_1496_to_fp16 = const()[name = tensor<string, []>("op_1496_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1497_cast = add(x = variance_33_cast, y = var_1496_to_fp16)[name = tensor<string, []>("op_1497_cast")];
            tensor<fp16, []> var_1498_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1498_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1498_cast = rsqrt(epsilon = var_1498_epsilon_0_to_fp16, x = var_1497_cast)[name = tensor<string, []>("op_1498_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_245_cast = mul(x = hidden_states_241_cast, y = var_1498_cast)[name = tensor<string, []>("hidden_states_245_cast")];
            tensor<fp16, [2048]> model_model_layers_8_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_8_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209034112)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_249_cast = mul(x = model_model_layers_8_input_layernorm_weight_to_fp16, y = hidden_states_245_cast)[name = tensor<string, []>("hidden_states_249_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_8_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209038272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211135488))), name = tensor<string, []>("model_model_layers_8_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_56_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_8_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_249_cast)[name = tensor<string, []>("linear_56_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_8_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211135616))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211397824))), name = tensor<string, []>("model_model_layers_8_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_57_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_8_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_249_cast)[name = tensor<string, []>("linear_57_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_8_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211397952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211660160))), name = tensor<string, []>("model_model_layers_8_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_58_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_8_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_249_cast)[name = tensor<string, []>("linear_58_cast")];
            tensor<int32, [4]> var_1518 = const()[name = tensor<string, []>("op_1518"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1519_cast = reshape(shape = var_1518, x = linear_56_cast)[name = tensor<string, []>("op_1519_cast")];
            tensor<int32, [4]> q_17_perm_0 = const()[name = tensor<string, []>("q_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1521 = const()[name = tensor<string, []>("op_1521"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1522_cast = reshape(shape = var_1521, x = linear_57_cast)[name = tensor<string, []>("op_1522_cast")];
            tensor<int32, [4]> key_states_51_perm_0 = const()[name = tensor<string, []>("key_states_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1524 = const()[name = tensor<string, []>("op_1524"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1525_cast = reshape(shape = var_1524, x = linear_58_cast)[name = tensor<string, []>("op_1525_cast")];
            tensor<int32, [4]> hidden_states_255_perm_0 = const()[name = tensor<string, []>("hidden_states_255_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_69 = transpose(perm = q_17_perm_0, x = var_1519_cast)[name = tensor<string, []>("transpose_69")];
            tensor<fp16, [1, 32, 128, 64]> var_1551_cast = mul(x = transpose_69, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1551_cast")];
            tensor<int32, [4]> x1_33_begin_0 = const()[name = tensor<string, []>("x1_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_33_end_0 = const()[name = tensor<string, []>("x1_33_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_33_end_mask_0 = const()[name = tensor<string, []>("x1_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_33_cast = slice_by_index(begin = x1_33_begin_0, end = x1_33_end_0, end_mask = x1_33_end_mask_0, x = transpose_69)[name = tensor<string, []>("x1_33_cast")];
            tensor<int32, [4]> x2_33_begin_0 = const()[name = tensor<string, []>("x2_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_33_end_0 = const()[name = tensor<string, []>("x2_33_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_33_end_mask_0 = const()[name = tensor<string, []>("x2_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_33_cast = slice_by_index(begin = x2_33_begin_0, end = x2_33_end_0, end_mask = x2_33_end_mask_0, x = transpose_69)[name = tensor<string, []>("x2_33_cast")];
            tensor<fp16, []> const_147_promoted_to_fp16 = const()[name = tensor<string, []>("const_147_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1562_cast = mul(x = x2_33_cast, y = const_147_promoted_to_fp16)[name = tensor<string, []>("op_1562_cast")];
            tensor<bool, []> var_1564_interleave_0 = const()[name = tensor<string, []>("op_1564_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1564_cast = concat(axis = var_10, interleave = var_1564_interleave_0, values = (var_1562_cast, x1_33_cast))[name = tensor<string, []>("op_1564_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1565_cast = mul(x = var_1564_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1565_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_35_cast = add(x = var_1551_cast, y = var_1565_cast)[name = tensor<string, []>("query_states_35_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_68 = transpose(perm = key_states_51_perm_0, x = var_1522_cast)[name = tensor<string, []>("transpose_68")];
            tensor<fp16, [1, 4, 128, 64]> var_1567_cast = mul(x = transpose_68, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1567_cast")];
            tensor<int32, [4]> x1_35_begin_0 = const()[name = tensor<string, []>("x1_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_35_end_0 = const()[name = tensor<string, []>("x1_35_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_35_end_mask_0 = const()[name = tensor<string, []>("x1_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_35_cast = slice_by_index(begin = x1_35_begin_0, end = x1_35_end_0, end_mask = x1_35_end_mask_0, x = transpose_68)[name = tensor<string, []>("x1_35_cast")];
            tensor<int32, [4]> x2_35_begin_0 = const()[name = tensor<string, []>("x2_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_35_end_0 = const()[name = tensor<string, []>("x2_35_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_35_end_mask_0 = const()[name = tensor<string, []>("x2_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_35_cast = slice_by_index(begin = x2_35_begin_0, end = x2_35_end_0, end_mask = x2_35_end_mask_0, x = transpose_68)[name = tensor<string, []>("x2_35_cast")];
            tensor<fp16, []> const_150_promoted_to_fp16 = const()[name = tensor<string, []>("const_150_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1578_cast = mul(x = x2_35_cast, y = const_150_promoted_to_fp16)[name = tensor<string, []>("op_1578_cast")];
            tensor<bool, []> var_1580_interleave_0 = const()[name = tensor<string, []>("op_1580_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1580_cast = concat(axis = var_10, interleave = var_1580_interleave_0, values = (var_1578_cast, x1_35_cast))[name = tensor<string, []>("op_1580_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1581_cast = mul(x = var_1580_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1581_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_251_cast = add(x = var_1567_cast, y = var_1581_cast)[name = tensor<string, []>("hidden_states_251_cast")];
            tensor<int32, [1]> var_1590_axes_0 = const()[name = tensor<string, []>("op_1590_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1590_cast = expand_dims(axes = var_1590_axes_0, x = hidden_states_251_cast)[name = tensor<string, []>("op_1590_cast")];
            tensor<int32, [5]> hidden_states_253_reps_0 = const()[name = tensor<string, []>("hidden_states_253_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_253_cast = tile(reps = hidden_states_253_reps_0, x = var_1590_cast)[name = tensor<string, []>("hidden_states_253_cast")];
            tensor<int32, [4]> var_1598 = const()[name = tensor<string, []>("op_1598"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_53_cast = reshape(shape = var_1598, x = hidden_states_253_cast)[name = tensor<string, []>("key_states_53_cast")];
            tensor<int32, [1]> var_1607_axes_0 = const()[name = tensor<string, []>("op_1607_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_67 = transpose(perm = hidden_states_255_perm_0, x = var_1525_cast)[name = tensor<string, []>("transpose_67")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1607_cast = expand_dims(axes = var_1607_axes_0, x = transpose_67)[name = tensor<string, []>("op_1607_cast")];
            tensor<int32, [5]> hidden_states_257_reps_0 = const()[name = tensor<string, []>("hidden_states_257_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_257_cast = tile(reps = hidden_states_257_reps_0, x = var_1607_cast)[name = tensor<string, []>("hidden_states_257_cast")];
            tensor<int32, [4]> var_1615 = const()[name = tensor<string, []>("op_1615"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_35_cast = reshape(shape = var_1615, x = hidden_states_257_cast)[name = tensor<string, []>("value_states_35_cast")];
            tensor<int32, [4]> var_1617_perm_0 = const()[name = tensor<string, []>("op_1617_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1618_transpose_x_0 = const()[name = tensor<string, []>("op_1618_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1618_transpose_y_0 = const()[name = tensor<string, []>("op_1618_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_66 = transpose(perm = var_1617_perm_0, x = key_states_53_cast)[name = tensor<string, []>("transpose_66")];
            tensor<fp16, [1, 32, 128, 128]> var_1618_cast = matmul(transpose_x = var_1618_transpose_x_0, transpose_y = var_1618_transpose_y_0, x = query_states_35_cast, y = transpose_66)[name = tensor<string, []>("op_1618_cast")];
            tensor<fp16, []> _inversed_attn_weights_33_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_33_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_33_cast = mul(x = var_1618_cast, y = _inversed_attn_weights_33_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_33_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_81_cast = add(x = _inversed_attn_weights_33_cast, y = attention_mask_cast)[name = tensor<string, []>("input_81_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1622_cast = softmax(axis = var_10, x = input_81_cast)[name = tensor<string, []>("op_1622_cast")];
            tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_33_cast = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = var_1622_cast, y = value_states_35_cast)[name = tensor<string, []>("attn_output_33_cast")];
            tensor<int32, [4]> var_1625_perm_0 = const()[name = tensor<string, []>("op_1625_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1627 = const()[name = tensor<string, []>("op_1627"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_65 = transpose(perm = var_1625_perm_0, x = attn_output_33_cast)[name = tensor<string, []>("transpose_65")];
            tensor<fp16, [1, 128, 2048]> input_83_cast = reshape(shape = var_1627, x = transpose_65)[name = tensor<string, []>("input_83_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_8_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211660288))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213757504))), name = tensor<string, []>("model_model_layers_8_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_59_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_8_self_attn_o_proj_weight_to_fp16_palettized, x = input_83_cast)[name = tensor<string, []>("linear_59_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_261_cast = add(x = hidden_states_241_cast, y = linear_59_cast)[name = tensor<string, []>("hidden_states_261_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_17 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_17"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1634_cast = pow(x = hidden_states_261_cast, y = var_13_promoted_to_fp16_17)[name = tensor<string, []>("op_1634_cast")];
            tensor<int32, [1]> var_1635 = const()[name = tensor<string, []>("op_1635"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_35_cast = reduce_mean(axes = var_1635, keep_dims = var_23, x = var_1634_cast)[name = tensor<string, []>("variance_35_cast")];
            tensor<fp16, []> var_1637_to_fp16 = const()[name = tensor<string, []>("op_1637_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1638_cast = add(x = variance_35_cast, y = var_1637_to_fp16)[name = tensor<string, []>("op_1638_cast")];
            tensor<fp16, []> var_1639_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1639_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1639_cast = rsqrt(epsilon = var_1639_epsilon_0_to_fp16, x = var_1638_cast)[name = tensor<string, []>("op_1639_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_265_cast = mul(x = hidden_states_261_cast, y = var_1639_cast)[name = tensor<string, []>("hidden_states_265_cast")];
            tensor<fp16, [2048]> model_model_layers_8_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_8_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213757632)))];
            tensor<fp16, [1, 128, 2048]> input_85_cast = mul(x = model_model_layers_8_post_attention_layernorm_weight_to_fp16, y = hidden_states_265_cast)[name = tensor<string, []>("input_85_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_8_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213761792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219529024))), name = tensor<string, []>("model_model_layers_8_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_60_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_8_mlp_gate_proj_weight_to_fp16_palettized, x = input_85_cast)[name = tensor<string, []>("linear_60_cast")];
            tensor<fp16, [1, 128, 5632]> var_1651_cast = silu(x = linear_60_cast)[name = tensor<string, []>("op_1651_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_8_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219529152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225296384))), name = tensor<string, []>("model_model_layers_8_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_61_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_8_mlp_up_proj_weight_to_fp16_palettized, x = input_85_cast)[name = tensor<string, []>("linear_61_cast")];
            tensor<fp16, [1, 128, 5632]> input_89_cast = mul(x = var_1651_cast, y = linear_61_cast)[name = tensor<string, []>("input_89_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_8_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225296512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231063744))), name = tensor<string, []>("model_model_layers_8_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_62_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_8_mlp_down_proj_weight_to_fp16_palettized, x = input_89_cast)[name = tensor<string, []>("linear_62_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_271_cast = add(x = hidden_states_261_cast, y = linear_62_cast)[name = tensor<string, []>("hidden_states_271_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_18 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_18"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1664_cast = pow(x = hidden_states_271_cast, y = var_13_promoted_to_fp16_18)[name = tensor<string, []>("op_1664_cast")];
            tensor<int32, [1]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_37_cast = reduce_mean(axes = var_1665, keep_dims = var_23, x = var_1664_cast)[name = tensor<string, []>("variance_37_cast")];
            tensor<fp16, []> var_1667_to_fp16 = const()[name = tensor<string, []>("op_1667_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1668_cast = add(x = variance_37_cast, y = var_1667_to_fp16)[name = tensor<string, []>("op_1668_cast")];
            tensor<fp16, []> var_1669_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1669_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1669_cast = rsqrt(epsilon = var_1669_epsilon_0_to_fp16, x = var_1668_cast)[name = tensor<string, []>("op_1669_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_275_cast = mul(x = hidden_states_271_cast, y = var_1669_cast)[name = tensor<string, []>("hidden_states_275_cast")];
            tensor<fp16, [2048]> model_model_layers_9_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_9_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231063872)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_279_cast = mul(x = model_model_layers_9_input_layernorm_weight_to_fp16, y = hidden_states_275_cast)[name = tensor<string, []>("hidden_states_279_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_9_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(231068032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233165248))), name = tensor<string, []>("model_model_layers_9_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_63_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_9_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_279_cast)[name = tensor<string, []>("linear_63_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_9_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233165376))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233427584))), name = tensor<string, []>("model_model_layers_9_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_64_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_9_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_279_cast)[name = tensor<string, []>("linear_64_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_9_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233427712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233689920))), name = tensor<string, []>("model_model_layers_9_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_65_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_9_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_279_cast)[name = tensor<string, []>("linear_65_cast")];
            tensor<int32, [4]> var_1689 = const()[name = tensor<string, []>("op_1689"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1690_cast = reshape(shape = var_1689, x = linear_63_cast)[name = tensor<string, []>("op_1690_cast")];
            tensor<int32, [4]> q_19_perm_0 = const()[name = tensor<string, []>("q_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1692 = const()[name = tensor<string, []>("op_1692"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1693_cast = reshape(shape = var_1692, x = linear_64_cast)[name = tensor<string, []>("op_1693_cast")];
            tensor<int32, [4]> key_states_57_perm_0 = const()[name = tensor<string, []>("key_states_57_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1695 = const()[name = tensor<string, []>("op_1695"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1696_cast = reshape(shape = var_1695, x = linear_65_cast)[name = tensor<string, []>("op_1696_cast")];
            tensor<int32, [4]> hidden_states_285_perm_0 = const()[name = tensor<string, []>("hidden_states_285_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_64 = transpose(perm = q_19_perm_0, x = var_1690_cast)[name = tensor<string, []>("transpose_64")];
            tensor<fp16, [1, 32, 128, 64]> var_1722_cast = mul(x = transpose_64, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1722_cast")];
            tensor<int32, [4]> x1_37_begin_0 = const()[name = tensor<string, []>("x1_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_37_end_0 = const()[name = tensor<string, []>("x1_37_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_37_end_mask_0 = const()[name = tensor<string, []>("x1_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_37_cast = slice_by_index(begin = x1_37_begin_0, end = x1_37_end_0, end_mask = x1_37_end_mask_0, x = transpose_64)[name = tensor<string, []>("x1_37_cast")];
            tensor<int32, [4]> x2_37_begin_0 = const()[name = tensor<string, []>("x2_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_37_end_0 = const()[name = tensor<string, []>("x2_37_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_37_end_mask_0 = const()[name = tensor<string, []>("x2_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_37_cast = slice_by_index(begin = x2_37_begin_0, end = x2_37_end_0, end_mask = x2_37_end_mask_0, x = transpose_64)[name = tensor<string, []>("x2_37_cast")];
            tensor<fp16, []> const_164_promoted_to_fp16 = const()[name = tensor<string, []>("const_164_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1733_cast = mul(x = x2_37_cast, y = const_164_promoted_to_fp16)[name = tensor<string, []>("op_1733_cast")];
            tensor<bool, []> var_1735_interleave_0 = const()[name = tensor<string, []>("op_1735_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1735_cast = concat(axis = var_10, interleave = var_1735_interleave_0, values = (var_1733_cast, x1_37_cast))[name = tensor<string, []>("op_1735_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1736_cast = mul(x = var_1735_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1736_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_39_cast = add(x = var_1722_cast, y = var_1736_cast)[name = tensor<string, []>("query_states_39_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_63 = transpose(perm = key_states_57_perm_0, x = var_1693_cast)[name = tensor<string, []>("transpose_63")];
            tensor<fp16, [1, 4, 128, 64]> var_1738_cast = mul(x = transpose_63, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1738_cast")];
            tensor<int32, [4]> x1_39_begin_0 = const()[name = tensor<string, []>("x1_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_39_end_0 = const()[name = tensor<string, []>("x1_39_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_39_end_mask_0 = const()[name = tensor<string, []>("x1_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_39_cast = slice_by_index(begin = x1_39_begin_0, end = x1_39_end_0, end_mask = x1_39_end_mask_0, x = transpose_63)[name = tensor<string, []>("x1_39_cast")];
            tensor<int32, [4]> x2_39_begin_0 = const()[name = tensor<string, []>("x2_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_39_end_0 = const()[name = tensor<string, []>("x2_39_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_39_end_mask_0 = const()[name = tensor<string, []>("x2_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_39_cast = slice_by_index(begin = x2_39_begin_0, end = x2_39_end_0, end_mask = x2_39_end_mask_0, x = transpose_63)[name = tensor<string, []>("x2_39_cast")];
            tensor<fp16, []> const_167_promoted_to_fp16 = const()[name = tensor<string, []>("const_167_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1749_cast = mul(x = x2_39_cast, y = const_167_promoted_to_fp16)[name = tensor<string, []>("op_1749_cast")];
            tensor<bool, []> var_1751_interleave_0 = const()[name = tensor<string, []>("op_1751_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1751_cast = concat(axis = var_10, interleave = var_1751_interleave_0, values = (var_1749_cast, x1_39_cast))[name = tensor<string, []>("op_1751_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1752_cast = mul(x = var_1751_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1752_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_281_cast = add(x = var_1738_cast, y = var_1752_cast)[name = tensor<string, []>("hidden_states_281_cast")];
            tensor<int32, [1]> var_1761_axes_0 = const()[name = tensor<string, []>("op_1761_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1761_cast = expand_dims(axes = var_1761_axes_0, x = hidden_states_281_cast)[name = tensor<string, []>("op_1761_cast")];
            tensor<int32, [5]> hidden_states_283_reps_0 = const()[name = tensor<string, []>("hidden_states_283_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_283_cast = tile(reps = hidden_states_283_reps_0, x = var_1761_cast)[name = tensor<string, []>("hidden_states_283_cast")];
            tensor<int32, [4]> var_1769 = const()[name = tensor<string, []>("op_1769"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_59_cast = reshape(shape = var_1769, x = hidden_states_283_cast)[name = tensor<string, []>("key_states_59_cast")];
            tensor<int32, [1]> var_1778_axes_0 = const()[name = tensor<string, []>("op_1778_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_62 = transpose(perm = hidden_states_285_perm_0, x = var_1696_cast)[name = tensor<string, []>("transpose_62")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1778_cast = expand_dims(axes = var_1778_axes_0, x = transpose_62)[name = tensor<string, []>("op_1778_cast")];
            tensor<int32, [5]> hidden_states_287_reps_0 = const()[name = tensor<string, []>("hidden_states_287_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_287_cast = tile(reps = hidden_states_287_reps_0, x = var_1778_cast)[name = tensor<string, []>("hidden_states_287_cast")];
            tensor<int32, [4]> var_1786 = const()[name = tensor<string, []>("op_1786"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_39_cast = reshape(shape = var_1786, x = hidden_states_287_cast)[name = tensor<string, []>("value_states_39_cast")];
            tensor<int32, [4]> var_1788_perm_0 = const()[name = tensor<string, []>("op_1788_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1789_transpose_x_0 = const()[name = tensor<string, []>("op_1789_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1789_transpose_y_0 = const()[name = tensor<string, []>("op_1789_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_61 = transpose(perm = var_1788_perm_0, x = key_states_59_cast)[name = tensor<string, []>("transpose_61")];
            tensor<fp16, [1, 32, 128, 128]> var_1789_cast = matmul(transpose_x = var_1789_transpose_x_0, transpose_y = var_1789_transpose_y_0, x = query_states_39_cast, y = transpose_61)[name = tensor<string, []>("op_1789_cast")];
            tensor<fp16, []> _inversed_attn_weights_37_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_37_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_37_cast = mul(x = var_1789_cast, y = _inversed_attn_weights_37_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_37_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_91_cast = add(x = _inversed_attn_weights_37_cast, y = attention_mask_cast)[name = tensor<string, []>("input_91_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1793_cast = softmax(axis = var_10, x = input_91_cast)[name = tensor<string, []>("op_1793_cast")];
            tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_37_cast = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = var_1793_cast, y = value_states_39_cast)[name = tensor<string, []>("attn_output_37_cast")];
            tensor<int32, [4]> var_1796_perm_0 = const()[name = tensor<string, []>("op_1796_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1798 = const()[name = tensor<string, []>("op_1798"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_60 = transpose(perm = var_1796_perm_0, x = attn_output_37_cast)[name = tensor<string, []>("transpose_60")];
            tensor<fp16, [1, 128, 2048]> input_93_cast = reshape(shape = var_1798, x = transpose_60)[name = tensor<string, []>("input_93_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_9_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233690048))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235787264))), name = tensor<string, []>("model_model_layers_9_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_66_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_9_self_attn_o_proj_weight_to_fp16_palettized, x = input_93_cast)[name = tensor<string, []>("linear_66_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_291_cast = add(x = hidden_states_271_cast, y = linear_66_cast)[name = tensor<string, []>("hidden_states_291_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_19 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_19"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1805_cast = pow(x = hidden_states_291_cast, y = var_13_promoted_to_fp16_19)[name = tensor<string, []>("op_1805_cast")];
            tensor<int32, [1]> var_1806 = const()[name = tensor<string, []>("op_1806"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_39_cast = reduce_mean(axes = var_1806, keep_dims = var_23, x = var_1805_cast)[name = tensor<string, []>("variance_39_cast")];
            tensor<fp16, []> var_1808_to_fp16 = const()[name = tensor<string, []>("op_1808_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1809_cast = add(x = variance_39_cast, y = var_1808_to_fp16)[name = tensor<string, []>("op_1809_cast")];
            tensor<fp16, []> var_1810_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1810_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1810_cast = rsqrt(epsilon = var_1810_epsilon_0_to_fp16, x = var_1809_cast)[name = tensor<string, []>("op_1810_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_295_cast = mul(x = hidden_states_291_cast, y = var_1810_cast)[name = tensor<string, []>("hidden_states_295_cast")];
            tensor<fp16, [2048]> model_model_layers_9_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_9_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235787392)))];
            tensor<fp16, [1, 128, 2048]> input_95_cast = mul(x = model_model_layers_9_post_attention_layernorm_weight_to_fp16, y = hidden_states_295_cast)[name = tensor<string, []>("input_95_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_9_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235791552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241558784))), name = tensor<string, []>("model_model_layers_9_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_67_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_9_mlp_gate_proj_weight_to_fp16_palettized, x = input_95_cast)[name = tensor<string, []>("linear_67_cast")];
            tensor<fp16, [1, 128, 5632]> var_1822_cast = silu(x = linear_67_cast)[name = tensor<string, []>("op_1822_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_9_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(241558912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(247326144))), name = tensor<string, []>("model_model_layers_9_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_68_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_9_mlp_up_proj_weight_to_fp16_palettized, x = input_95_cast)[name = tensor<string, []>("linear_68_cast")];
            tensor<fp16, [1, 128, 5632]> input_99_cast = mul(x = var_1822_cast, y = linear_68_cast)[name = tensor<string, []>("input_99_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_9_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(247326272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253093504))), name = tensor<string, []>("model_model_layers_9_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_69_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_9_mlp_down_proj_weight_to_fp16_palettized, x = input_99_cast)[name = tensor<string, []>("linear_69_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_301_cast = add(x = hidden_states_291_cast, y = linear_69_cast)[name = tensor<string, []>("hidden_states_301_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_20 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_20"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1835_cast = pow(x = hidden_states_301_cast, y = var_13_promoted_to_fp16_20)[name = tensor<string, []>("op_1835_cast")];
            tensor<int32, [1]> var_1836 = const()[name = tensor<string, []>("op_1836"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_41_cast = reduce_mean(axes = var_1836, keep_dims = var_23, x = var_1835_cast)[name = tensor<string, []>("variance_41_cast")];
            tensor<fp16, []> var_1838_to_fp16 = const()[name = tensor<string, []>("op_1838_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1839_cast = add(x = variance_41_cast, y = var_1838_to_fp16)[name = tensor<string, []>("op_1839_cast")];
            tensor<fp16, []> var_1840_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1840_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1840_cast = rsqrt(epsilon = var_1840_epsilon_0_to_fp16, x = var_1839_cast)[name = tensor<string, []>("op_1840_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_305_cast = mul(x = hidden_states_301_cast, y = var_1840_cast)[name = tensor<string, []>("hidden_states_305_cast")];
            tensor<fp16, [2048]> model_model_layers_10_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_10_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253093632)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_309_cast = mul(x = model_model_layers_10_input_layernorm_weight_to_fp16, y = hidden_states_305_cast)[name = tensor<string, []>("hidden_states_309_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_10_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253097792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255195008))), name = tensor<string, []>("model_model_layers_10_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_70_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_10_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_309_cast)[name = tensor<string, []>("linear_70_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_10_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255195136))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255457344))), name = tensor<string, []>("model_model_layers_10_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_71_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_10_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_309_cast)[name = tensor<string, []>("linear_71_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_10_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255457472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255719680))), name = tensor<string, []>("model_model_layers_10_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_72_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_10_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_309_cast)[name = tensor<string, []>("linear_72_cast")];
            tensor<int32, [4]> var_1860 = const()[name = tensor<string, []>("op_1860"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_1861_cast = reshape(shape = var_1860, x = linear_70_cast)[name = tensor<string, []>("op_1861_cast")];
            tensor<int32, [4]> q_21_perm_0 = const()[name = tensor<string, []>("q_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1863 = const()[name = tensor<string, []>("op_1863"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1864_cast = reshape(shape = var_1863, x = linear_71_cast)[name = tensor<string, []>("op_1864_cast")];
            tensor<int32, [4]> key_states_63_perm_0 = const()[name = tensor<string, []>("key_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_1866 = const()[name = tensor<string, []>("op_1866"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_1867_cast = reshape(shape = var_1866, x = linear_72_cast)[name = tensor<string, []>("op_1867_cast")];
            tensor<int32, [4]> hidden_states_315_perm_0 = const()[name = tensor<string, []>("hidden_states_315_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_59 = transpose(perm = q_21_perm_0, x = var_1861_cast)[name = tensor<string, []>("transpose_59")];
            tensor<fp16, [1, 32, 128, 64]> var_1893_cast = mul(x = transpose_59, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1893_cast")];
            tensor<int32, [4]> x1_41_begin_0 = const()[name = tensor<string, []>("x1_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_41_end_0 = const()[name = tensor<string, []>("x1_41_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_41_end_mask_0 = const()[name = tensor<string, []>("x1_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_41_cast = slice_by_index(begin = x1_41_begin_0, end = x1_41_end_0, end_mask = x1_41_end_mask_0, x = transpose_59)[name = tensor<string, []>("x1_41_cast")];
            tensor<int32, [4]> x2_41_begin_0 = const()[name = tensor<string, []>("x2_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_41_end_0 = const()[name = tensor<string, []>("x2_41_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_41_end_mask_0 = const()[name = tensor<string, []>("x2_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_41_cast = slice_by_index(begin = x2_41_begin_0, end = x2_41_end_0, end_mask = x2_41_end_mask_0, x = transpose_59)[name = tensor<string, []>("x2_41_cast")];
            tensor<fp16, []> const_181_promoted_to_fp16 = const()[name = tensor<string, []>("const_181_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_1904_cast = mul(x = x2_41_cast, y = const_181_promoted_to_fp16)[name = tensor<string, []>("op_1904_cast")];
            tensor<bool, []> var_1906_interleave_0 = const()[name = tensor<string, []>("op_1906_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_1906_cast = concat(axis = var_10, interleave = var_1906_interleave_0, values = (var_1904_cast, x1_41_cast))[name = tensor<string, []>("op_1906_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_1907_cast = mul(x = var_1906_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1907_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_43_cast = add(x = var_1893_cast, y = var_1907_cast)[name = tensor<string, []>("query_states_43_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_58 = transpose(perm = key_states_63_perm_0, x = var_1864_cast)[name = tensor<string, []>("transpose_58")];
            tensor<fp16, [1, 4, 128, 64]> var_1909_cast = mul(x = transpose_58, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_1909_cast")];
            tensor<int32, [4]> x1_43_begin_0 = const()[name = tensor<string, []>("x1_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_43_end_0 = const()[name = tensor<string, []>("x1_43_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_43_end_mask_0 = const()[name = tensor<string, []>("x1_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_43_cast = slice_by_index(begin = x1_43_begin_0, end = x1_43_end_0, end_mask = x1_43_end_mask_0, x = transpose_58)[name = tensor<string, []>("x1_43_cast")];
            tensor<int32, [4]> x2_43_begin_0 = const()[name = tensor<string, []>("x2_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_43_end_0 = const()[name = tensor<string, []>("x2_43_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_43_end_mask_0 = const()[name = tensor<string, []>("x2_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_43_cast = slice_by_index(begin = x2_43_begin_0, end = x2_43_end_0, end_mask = x2_43_end_mask_0, x = transpose_58)[name = tensor<string, []>("x2_43_cast")];
            tensor<fp16, []> const_184_promoted_to_fp16 = const()[name = tensor<string, []>("const_184_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_1920_cast = mul(x = x2_43_cast, y = const_184_promoted_to_fp16)[name = tensor<string, []>("op_1920_cast")];
            tensor<bool, []> var_1922_interleave_0 = const()[name = tensor<string, []>("op_1922_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_1922_cast = concat(axis = var_10, interleave = var_1922_interleave_0, values = (var_1920_cast, x1_43_cast))[name = tensor<string, []>("op_1922_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_1923_cast = mul(x = var_1922_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_1923_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_311_cast = add(x = var_1909_cast, y = var_1923_cast)[name = tensor<string, []>("hidden_states_311_cast")];
            tensor<int32, [1]> var_1932_axes_0 = const()[name = tensor<string, []>("op_1932_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1932_cast = expand_dims(axes = var_1932_axes_0, x = hidden_states_311_cast)[name = tensor<string, []>("op_1932_cast")];
            tensor<int32, [5]> hidden_states_313_reps_0 = const()[name = tensor<string, []>("hidden_states_313_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_313_cast = tile(reps = hidden_states_313_reps_0, x = var_1932_cast)[name = tensor<string, []>("hidden_states_313_cast")];
            tensor<int32, [4]> var_1940 = const()[name = tensor<string, []>("op_1940"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_65_cast = reshape(shape = var_1940, x = hidden_states_313_cast)[name = tensor<string, []>("key_states_65_cast")];
            tensor<int32, [1]> var_1949_axes_0 = const()[name = tensor<string, []>("op_1949_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_57 = transpose(perm = hidden_states_315_perm_0, x = var_1867_cast)[name = tensor<string, []>("transpose_57")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_1949_cast = expand_dims(axes = var_1949_axes_0, x = transpose_57)[name = tensor<string, []>("op_1949_cast")];
            tensor<int32, [5]> hidden_states_317_reps_0 = const()[name = tensor<string, []>("hidden_states_317_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_317_cast = tile(reps = hidden_states_317_reps_0, x = var_1949_cast)[name = tensor<string, []>("hidden_states_317_cast")];
            tensor<int32, [4]> var_1957 = const()[name = tensor<string, []>("op_1957"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_43_cast = reshape(shape = var_1957, x = hidden_states_317_cast)[name = tensor<string, []>("value_states_43_cast")];
            tensor<int32, [4]> var_1959_perm_0 = const()[name = tensor<string, []>("op_1959_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_1960_transpose_x_0 = const()[name = tensor<string, []>("op_1960_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1960_transpose_y_0 = const()[name = tensor<string, []>("op_1960_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_56 = transpose(perm = var_1959_perm_0, x = key_states_65_cast)[name = tensor<string, []>("transpose_56")];
            tensor<fp16, [1, 32, 128, 128]> var_1960_cast = matmul(transpose_x = var_1960_transpose_x_0, transpose_y = var_1960_transpose_y_0, x = query_states_43_cast, y = transpose_56)[name = tensor<string, []>("op_1960_cast")];
            tensor<fp16, []> _inversed_attn_weights_41_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_41_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_41_cast = mul(x = var_1960_cast, y = _inversed_attn_weights_41_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_41_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_101_cast = add(x = _inversed_attn_weights_41_cast, y = attention_mask_cast)[name = tensor<string, []>("input_101_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_1964_cast = softmax(axis = var_10, x = input_101_cast)[name = tensor<string, []>("op_1964_cast")];
            tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_41_cast = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = var_1964_cast, y = value_states_43_cast)[name = tensor<string, []>("attn_output_41_cast")];
            tensor<int32, [4]> var_1967_perm_0 = const()[name = tensor<string, []>("op_1967_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1969 = const()[name = tensor<string, []>("op_1969"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_55 = transpose(perm = var_1967_perm_0, x = attn_output_41_cast)[name = tensor<string, []>("transpose_55")];
            tensor<fp16, [1, 128, 2048]> input_103_cast = reshape(shape = var_1969, x = transpose_55)[name = tensor<string, []>("input_103_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_10_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(255719808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257817024))), name = tensor<string, []>("model_model_layers_10_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_73_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_10_self_attn_o_proj_weight_to_fp16_palettized, x = input_103_cast)[name = tensor<string, []>("linear_73_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_321_cast = add(x = hidden_states_301_cast, y = linear_73_cast)[name = tensor<string, []>("hidden_states_321_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_21 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_21"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_1976_cast = pow(x = hidden_states_321_cast, y = var_13_promoted_to_fp16_21)[name = tensor<string, []>("op_1976_cast")];
            tensor<int32, [1]> var_1977 = const()[name = tensor<string, []>("op_1977"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_43_cast = reduce_mean(axes = var_1977, keep_dims = var_23, x = var_1976_cast)[name = tensor<string, []>("variance_43_cast")];
            tensor<fp16, []> var_1979_to_fp16 = const()[name = tensor<string, []>("op_1979_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_1980_cast = add(x = variance_43_cast, y = var_1979_to_fp16)[name = tensor<string, []>("op_1980_cast")];
            tensor<fp16, []> var_1981_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_1981_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_1981_cast = rsqrt(epsilon = var_1981_epsilon_0_to_fp16, x = var_1980_cast)[name = tensor<string, []>("op_1981_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_325_cast = mul(x = hidden_states_321_cast, y = var_1981_cast)[name = tensor<string, []>("hidden_states_325_cast")];
            tensor<fp16, [2048]> model_model_layers_10_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_10_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257817152)))];
            tensor<fp16, [1, 128, 2048]> input_105_cast = mul(x = model_model_layers_10_post_attention_layernorm_weight_to_fp16, y = hidden_states_325_cast)[name = tensor<string, []>("input_105_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_10_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(257821312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263588544))), name = tensor<string, []>("model_model_layers_10_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_74_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_10_mlp_gate_proj_weight_to_fp16_palettized, x = input_105_cast)[name = tensor<string, []>("linear_74_cast")];
            tensor<fp16, [1, 128, 5632]> var_1993_cast = silu(x = linear_74_cast)[name = tensor<string, []>("op_1993_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_10_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263588672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269355904))), name = tensor<string, []>("model_model_layers_10_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_75_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_10_mlp_up_proj_weight_to_fp16_palettized, x = input_105_cast)[name = tensor<string, []>("linear_75_cast")];
            tensor<fp16, [1, 128, 5632]> input_109_cast = mul(x = var_1993_cast, y = linear_75_cast)[name = tensor<string, []>("input_109_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_10_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269356032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275123264))), name = tensor<string, []>("model_model_layers_10_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_76_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_10_mlp_down_proj_weight_to_fp16_palettized, x = input_109_cast)[name = tensor<string, []>("linear_76_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_331_cast = add(x = hidden_states_321_cast, y = linear_76_cast)[name = tensor<string, []>("hidden_states_331_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_22 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_22"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2006_cast = pow(x = hidden_states_331_cast, y = var_13_promoted_to_fp16_22)[name = tensor<string, []>("op_2006_cast")];
            tensor<int32, [1]> var_2007 = const()[name = tensor<string, []>("op_2007"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_45_cast = reduce_mean(axes = var_2007, keep_dims = var_23, x = var_2006_cast)[name = tensor<string, []>("variance_45_cast")];
            tensor<fp16, []> var_2009_to_fp16 = const()[name = tensor<string, []>("op_2009_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2010_cast = add(x = variance_45_cast, y = var_2009_to_fp16)[name = tensor<string, []>("op_2010_cast")];
            tensor<fp16, []> var_2011_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2011_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2011_cast = rsqrt(epsilon = var_2011_epsilon_0_to_fp16, x = var_2010_cast)[name = tensor<string, []>("op_2011_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_335_cast = mul(x = hidden_states_331_cast, y = var_2011_cast)[name = tensor<string, []>("hidden_states_335_cast")];
            tensor<fp16, [2048]> model_model_layers_11_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_11_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275123392)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_339_cast = mul(x = model_model_layers_11_input_layernorm_weight_to_fp16, y = hidden_states_335_cast)[name = tensor<string, []>("hidden_states_339_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_11_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275127552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277224768))), name = tensor<string, []>("model_model_layers_11_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_77_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_11_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_339_cast)[name = tensor<string, []>("linear_77_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_11_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277224896))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277487104))), name = tensor<string, []>("model_model_layers_11_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_78_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_11_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_339_cast)[name = tensor<string, []>("linear_78_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_11_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277487232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277749440))), name = tensor<string, []>("model_model_layers_11_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_79_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_11_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_339_cast)[name = tensor<string, []>("linear_79_cast")];
            tensor<int32, [4]> var_2031 = const()[name = tensor<string, []>("op_2031"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2032_cast = reshape(shape = var_2031, x = linear_77_cast)[name = tensor<string, []>("op_2032_cast")];
            tensor<int32, [4]> q_23_perm_0 = const()[name = tensor<string, []>("q_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2034 = const()[name = tensor<string, []>("op_2034"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2035_cast = reshape(shape = var_2034, x = linear_78_cast)[name = tensor<string, []>("op_2035_cast")];
            tensor<int32, [4]> key_states_69_perm_0 = const()[name = tensor<string, []>("key_states_69_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2037 = const()[name = tensor<string, []>("op_2037"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2038_cast = reshape(shape = var_2037, x = linear_79_cast)[name = tensor<string, []>("op_2038_cast")];
            tensor<int32, [4]> hidden_states_345_perm_0 = const()[name = tensor<string, []>("hidden_states_345_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_54 = transpose(perm = q_23_perm_0, x = var_2032_cast)[name = tensor<string, []>("transpose_54")];
            tensor<fp16, [1, 32, 128, 64]> var_2064_cast = mul(x = transpose_54, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2064_cast")];
            tensor<int32, [4]> x1_45_begin_0 = const()[name = tensor<string, []>("x1_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_45_end_0 = const()[name = tensor<string, []>("x1_45_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_45_end_mask_0 = const()[name = tensor<string, []>("x1_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_45_cast = slice_by_index(begin = x1_45_begin_0, end = x1_45_end_0, end_mask = x1_45_end_mask_0, x = transpose_54)[name = tensor<string, []>("x1_45_cast")];
            tensor<int32, [4]> x2_45_begin_0 = const()[name = tensor<string, []>("x2_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_45_end_0 = const()[name = tensor<string, []>("x2_45_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_45_end_mask_0 = const()[name = tensor<string, []>("x2_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_45_cast = slice_by_index(begin = x2_45_begin_0, end = x2_45_end_0, end_mask = x2_45_end_mask_0, x = transpose_54)[name = tensor<string, []>("x2_45_cast")];
            tensor<fp16, []> const_198_promoted_to_fp16 = const()[name = tensor<string, []>("const_198_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2075_cast = mul(x = x2_45_cast, y = const_198_promoted_to_fp16)[name = tensor<string, []>("op_2075_cast")];
            tensor<bool, []> var_2077_interleave_0 = const()[name = tensor<string, []>("op_2077_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2077_cast = concat(axis = var_10, interleave = var_2077_interleave_0, values = (var_2075_cast, x1_45_cast))[name = tensor<string, []>("op_2077_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2078_cast = mul(x = var_2077_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2078_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_47_cast = add(x = var_2064_cast, y = var_2078_cast)[name = tensor<string, []>("query_states_47_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_53 = transpose(perm = key_states_69_perm_0, x = var_2035_cast)[name = tensor<string, []>("transpose_53")];
            tensor<fp16, [1, 4, 128, 64]> var_2080_cast = mul(x = transpose_53, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2080_cast")];
            tensor<int32, [4]> x1_47_begin_0 = const()[name = tensor<string, []>("x1_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_47_end_0 = const()[name = tensor<string, []>("x1_47_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_47_end_mask_0 = const()[name = tensor<string, []>("x1_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_47_cast = slice_by_index(begin = x1_47_begin_0, end = x1_47_end_0, end_mask = x1_47_end_mask_0, x = transpose_53)[name = tensor<string, []>("x1_47_cast")];
            tensor<int32, [4]> x2_47_begin_0 = const()[name = tensor<string, []>("x2_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_47_end_0 = const()[name = tensor<string, []>("x2_47_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_47_end_mask_0 = const()[name = tensor<string, []>("x2_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_47_cast = slice_by_index(begin = x2_47_begin_0, end = x2_47_end_0, end_mask = x2_47_end_mask_0, x = transpose_53)[name = tensor<string, []>("x2_47_cast")];
            tensor<fp16, []> const_201_promoted_to_fp16 = const()[name = tensor<string, []>("const_201_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2091_cast = mul(x = x2_47_cast, y = const_201_promoted_to_fp16)[name = tensor<string, []>("op_2091_cast")];
            tensor<bool, []> var_2093_interleave_0 = const()[name = tensor<string, []>("op_2093_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2093_cast = concat(axis = var_10, interleave = var_2093_interleave_0, values = (var_2091_cast, x1_47_cast))[name = tensor<string, []>("op_2093_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2094_cast = mul(x = var_2093_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2094_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_341_cast = add(x = var_2080_cast, y = var_2094_cast)[name = tensor<string, []>("hidden_states_341_cast")];
            tensor<int32, [1]> var_2103_axes_0 = const()[name = tensor<string, []>("op_2103_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2103_cast = expand_dims(axes = var_2103_axes_0, x = hidden_states_341_cast)[name = tensor<string, []>("op_2103_cast")];
            tensor<int32, [5]> hidden_states_343_reps_0 = const()[name = tensor<string, []>("hidden_states_343_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_343_cast = tile(reps = hidden_states_343_reps_0, x = var_2103_cast)[name = tensor<string, []>("hidden_states_343_cast")];
            tensor<int32, [4]> var_2111 = const()[name = tensor<string, []>("op_2111"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_71_cast = reshape(shape = var_2111, x = hidden_states_343_cast)[name = tensor<string, []>("key_states_71_cast")];
            tensor<int32, [1]> var_2120_axes_0 = const()[name = tensor<string, []>("op_2120_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_52 = transpose(perm = hidden_states_345_perm_0, x = var_2038_cast)[name = tensor<string, []>("transpose_52")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2120_cast = expand_dims(axes = var_2120_axes_0, x = transpose_52)[name = tensor<string, []>("op_2120_cast")];
            tensor<int32, [5]> hidden_states_347_reps_0 = const()[name = tensor<string, []>("hidden_states_347_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_347_cast = tile(reps = hidden_states_347_reps_0, x = var_2120_cast)[name = tensor<string, []>("hidden_states_347_cast")];
            tensor<int32, [4]> var_2128 = const()[name = tensor<string, []>("op_2128"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_47_cast = reshape(shape = var_2128, x = hidden_states_347_cast)[name = tensor<string, []>("value_states_47_cast")];
            tensor<int32, [4]> var_2130_perm_0 = const()[name = tensor<string, []>("op_2130_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2131_transpose_x_0 = const()[name = tensor<string, []>("op_2131_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2131_transpose_y_0 = const()[name = tensor<string, []>("op_2131_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_51 = transpose(perm = var_2130_perm_0, x = key_states_71_cast)[name = tensor<string, []>("transpose_51")];
            tensor<fp16, [1, 32, 128, 128]> var_2131_cast = matmul(transpose_x = var_2131_transpose_x_0, transpose_y = var_2131_transpose_y_0, x = query_states_47_cast, y = transpose_51)[name = tensor<string, []>("op_2131_cast")];
            tensor<fp16, []> _inversed_attn_weights_45_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_45_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_45_cast = mul(x = var_2131_cast, y = _inversed_attn_weights_45_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_45_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_111_cast = add(x = _inversed_attn_weights_45_cast, y = attention_mask_cast)[name = tensor<string, []>("input_111_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2135_cast = softmax(axis = var_10, x = input_111_cast)[name = tensor<string, []>("op_2135_cast")];
            tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_45_cast = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = var_2135_cast, y = value_states_47_cast)[name = tensor<string, []>("attn_output_45_cast")];
            tensor<int32, [4]> var_2138_perm_0 = const()[name = tensor<string, []>("op_2138_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2140 = const()[name = tensor<string, []>("op_2140"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_50 = transpose(perm = var_2138_perm_0, x = attn_output_45_cast)[name = tensor<string, []>("transpose_50")];
            tensor<fp16, [1, 128, 2048]> input_113_cast = reshape(shape = var_2140, x = transpose_50)[name = tensor<string, []>("input_113_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_11_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(277749568))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279846784))), name = tensor<string, []>("model_model_layers_11_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_80_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_11_self_attn_o_proj_weight_to_fp16_palettized, x = input_113_cast)[name = tensor<string, []>("linear_80_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_351_cast = add(x = hidden_states_331_cast, y = linear_80_cast)[name = tensor<string, []>("hidden_states_351_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_23 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_23"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2147_cast = pow(x = hidden_states_351_cast, y = var_13_promoted_to_fp16_23)[name = tensor<string, []>("op_2147_cast")];
            tensor<int32, [1]> var_2148 = const()[name = tensor<string, []>("op_2148"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_47_cast = reduce_mean(axes = var_2148, keep_dims = var_23, x = var_2147_cast)[name = tensor<string, []>("variance_47_cast")];
            tensor<fp16, []> var_2150_to_fp16 = const()[name = tensor<string, []>("op_2150_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2151_cast = add(x = variance_47_cast, y = var_2150_to_fp16)[name = tensor<string, []>("op_2151_cast")];
            tensor<fp16, []> var_2152_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2152_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2152_cast = rsqrt(epsilon = var_2152_epsilon_0_to_fp16, x = var_2151_cast)[name = tensor<string, []>("op_2152_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_355_cast = mul(x = hidden_states_351_cast, y = var_2152_cast)[name = tensor<string, []>("hidden_states_355_cast")];
            tensor<fp16, [2048]> model_model_layers_11_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_11_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279846912)))];
            tensor<fp16, [1, 128, 2048]> input_115_cast = mul(x = model_model_layers_11_post_attention_layernorm_weight_to_fp16, y = hidden_states_355_cast)[name = tensor<string, []>("input_115_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_11_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279851072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(285618304))), name = tensor<string, []>("model_model_layers_11_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_81_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_11_mlp_gate_proj_weight_to_fp16_palettized, x = input_115_cast)[name = tensor<string, []>("linear_81_cast")];
            tensor<fp16, [1, 128, 5632]> var_2164_cast = silu(x = linear_81_cast)[name = tensor<string, []>("op_2164_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_11_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(285618432))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291385664))), name = tensor<string, []>("model_model_layers_11_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_82_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_11_mlp_up_proj_weight_to_fp16_palettized, x = input_115_cast)[name = tensor<string, []>("linear_82_cast")];
            tensor<fp16, [1, 128, 5632]> input_119_cast = mul(x = var_2164_cast, y = linear_82_cast)[name = tensor<string, []>("input_119_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_11_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(291385792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297153024))), name = tensor<string, []>("model_model_layers_11_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_83_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_11_mlp_down_proj_weight_to_fp16_palettized, x = input_119_cast)[name = tensor<string, []>("linear_83_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_361_cast = add(x = hidden_states_351_cast, y = linear_83_cast)[name = tensor<string, []>("hidden_states_361_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_24 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_24"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2177_cast = pow(x = hidden_states_361_cast, y = var_13_promoted_to_fp16_24)[name = tensor<string, []>("op_2177_cast")];
            tensor<int32, [1]> var_2178 = const()[name = tensor<string, []>("op_2178"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_49_cast = reduce_mean(axes = var_2178, keep_dims = var_23, x = var_2177_cast)[name = tensor<string, []>("variance_49_cast")];
            tensor<fp16, []> var_2180_to_fp16 = const()[name = tensor<string, []>("op_2180_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2181_cast = add(x = variance_49_cast, y = var_2180_to_fp16)[name = tensor<string, []>("op_2181_cast")];
            tensor<fp16, []> var_2182_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2182_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2182_cast = rsqrt(epsilon = var_2182_epsilon_0_to_fp16, x = var_2181_cast)[name = tensor<string, []>("op_2182_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_365_cast = mul(x = hidden_states_361_cast, y = var_2182_cast)[name = tensor<string, []>("hidden_states_365_cast")];
            tensor<fp16, [2048]> model_model_layers_12_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_12_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297153152)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_369_cast = mul(x = model_model_layers_12_input_layernorm_weight_to_fp16, y = hidden_states_365_cast)[name = tensor<string, []>("hidden_states_369_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_12_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(297157312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299254528))), name = tensor<string, []>("model_model_layers_12_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_84_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_12_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_369_cast)[name = tensor<string, []>("linear_84_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_12_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299254656))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299516864))), name = tensor<string, []>("model_model_layers_12_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_85_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_12_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_369_cast)[name = tensor<string, []>("linear_85_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_12_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299516992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299779200))), name = tensor<string, []>("model_model_layers_12_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_86_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_12_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_369_cast)[name = tensor<string, []>("linear_86_cast")];
            tensor<int32, [4]> var_2202 = const()[name = tensor<string, []>("op_2202"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2203_cast = reshape(shape = var_2202, x = linear_84_cast)[name = tensor<string, []>("op_2203_cast")];
            tensor<int32, [4]> q_25_perm_0 = const()[name = tensor<string, []>("q_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2205 = const()[name = tensor<string, []>("op_2205"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2206_cast = reshape(shape = var_2205, x = linear_85_cast)[name = tensor<string, []>("op_2206_cast")];
            tensor<int32, [4]> key_states_75_perm_0 = const()[name = tensor<string, []>("key_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2208 = const()[name = tensor<string, []>("op_2208"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2209_cast = reshape(shape = var_2208, x = linear_86_cast)[name = tensor<string, []>("op_2209_cast")];
            tensor<int32, [4]> hidden_states_375_perm_0 = const()[name = tensor<string, []>("hidden_states_375_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_49 = transpose(perm = q_25_perm_0, x = var_2203_cast)[name = tensor<string, []>("transpose_49")];
            tensor<fp16, [1, 32, 128, 64]> var_2235_cast = mul(x = transpose_49, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2235_cast")];
            tensor<int32, [4]> x1_49_begin_0 = const()[name = tensor<string, []>("x1_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_49_end_0 = const()[name = tensor<string, []>("x1_49_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_49_end_mask_0 = const()[name = tensor<string, []>("x1_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_49_cast = slice_by_index(begin = x1_49_begin_0, end = x1_49_end_0, end_mask = x1_49_end_mask_0, x = transpose_49)[name = tensor<string, []>("x1_49_cast")];
            tensor<int32, [4]> x2_49_begin_0 = const()[name = tensor<string, []>("x2_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_49_end_0 = const()[name = tensor<string, []>("x2_49_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_49_end_mask_0 = const()[name = tensor<string, []>("x2_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_49_cast = slice_by_index(begin = x2_49_begin_0, end = x2_49_end_0, end_mask = x2_49_end_mask_0, x = transpose_49)[name = tensor<string, []>("x2_49_cast")];
            tensor<fp16, []> const_215_promoted_to_fp16 = const()[name = tensor<string, []>("const_215_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2246_cast = mul(x = x2_49_cast, y = const_215_promoted_to_fp16)[name = tensor<string, []>("op_2246_cast")];
            tensor<bool, []> var_2248_interleave_0 = const()[name = tensor<string, []>("op_2248_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2248_cast = concat(axis = var_10, interleave = var_2248_interleave_0, values = (var_2246_cast, x1_49_cast))[name = tensor<string, []>("op_2248_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2249_cast = mul(x = var_2248_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2249_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_51_cast = add(x = var_2235_cast, y = var_2249_cast)[name = tensor<string, []>("query_states_51_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_48 = transpose(perm = key_states_75_perm_0, x = var_2206_cast)[name = tensor<string, []>("transpose_48")];
            tensor<fp16, [1, 4, 128, 64]> var_2251_cast = mul(x = transpose_48, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2251_cast")];
            tensor<int32, [4]> x1_51_begin_0 = const()[name = tensor<string, []>("x1_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_51_end_0 = const()[name = tensor<string, []>("x1_51_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_51_end_mask_0 = const()[name = tensor<string, []>("x1_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_51_cast = slice_by_index(begin = x1_51_begin_0, end = x1_51_end_0, end_mask = x1_51_end_mask_0, x = transpose_48)[name = tensor<string, []>("x1_51_cast")];
            tensor<int32, [4]> x2_51_begin_0 = const()[name = tensor<string, []>("x2_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_51_end_0 = const()[name = tensor<string, []>("x2_51_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_51_end_mask_0 = const()[name = tensor<string, []>("x2_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_51_cast = slice_by_index(begin = x2_51_begin_0, end = x2_51_end_0, end_mask = x2_51_end_mask_0, x = transpose_48)[name = tensor<string, []>("x2_51_cast")];
            tensor<fp16, []> const_218_promoted_to_fp16 = const()[name = tensor<string, []>("const_218_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2262_cast = mul(x = x2_51_cast, y = const_218_promoted_to_fp16)[name = tensor<string, []>("op_2262_cast")];
            tensor<bool, []> var_2264_interleave_0 = const()[name = tensor<string, []>("op_2264_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2264_cast = concat(axis = var_10, interleave = var_2264_interleave_0, values = (var_2262_cast, x1_51_cast))[name = tensor<string, []>("op_2264_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2265_cast = mul(x = var_2264_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2265_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_371_cast = add(x = var_2251_cast, y = var_2265_cast)[name = tensor<string, []>("hidden_states_371_cast")];
            tensor<int32, [1]> var_2274_axes_0 = const()[name = tensor<string, []>("op_2274_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2274_cast = expand_dims(axes = var_2274_axes_0, x = hidden_states_371_cast)[name = tensor<string, []>("op_2274_cast")];
            tensor<int32, [5]> hidden_states_373_reps_0 = const()[name = tensor<string, []>("hidden_states_373_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_373_cast = tile(reps = hidden_states_373_reps_0, x = var_2274_cast)[name = tensor<string, []>("hidden_states_373_cast")];
            tensor<int32, [4]> var_2282 = const()[name = tensor<string, []>("op_2282"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_77_cast = reshape(shape = var_2282, x = hidden_states_373_cast)[name = tensor<string, []>("key_states_77_cast")];
            tensor<int32, [1]> var_2291_axes_0 = const()[name = tensor<string, []>("op_2291_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_47 = transpose(perm = hidden_states_375_perm_0, x = var_2209_cast)[name = tensor<string, []>("transpose_47")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2291_cast = expand_dims(axes = var_2291_axes_0, x = transpose_47)[name = tensor<string, []>("op_2291_cast")];
            tensor<int32, [5]> hidden_states_377_reps_0 = const()[name = tensor<string, []>("hidden_states_377_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_377_cast = tile(reps = hidden_states_377_reps_0, x = var_2291_cast)[name = tensor<string, []>("hidden_states_377_cast")];
            tensor<int32, [4]> var_2299 = const()[name = tensor<string, []>("op_2299"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_51_cast = reshape(shape = var_2299, x = hidden_states_377_cast)[name = tensor<string, []>("value_states_51_cast")];
            tensor<int32, [4]> var_2301_perm_0 = const()[name = tensor<string, []>("op_2301_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2302_transpose_x_0 = const()[name = tensor<string, []>("op_2302_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2302_transpose_y_0 = const()[name = tensor<string, []>("op_2302_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_46 = transpose(perm = var_2301_perm_0, x = key_states_77_cast)[name = tensor<string, []>("transpose_46")];
            tensor<fp16, [1, 32, 128, 128]> var_2302_cast = matmul(transpose_x = var_2302_transpose_x_0, transpose_y = var_2302_transpose_y_0, x = query_states_51_cast, y = transpose_46)[name = tensor<string, []>("op_2302_cast")];
            tensor<fp16, []> _inversed_attn_weights_49_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_49_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_49_cast = mul(x = var_2302_cast, y = _inversed_attn_weights_49_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_49_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_121_cast = add(x = _inversed_attn_weights_49_cast, y = attention_mask_cast)[name = tensor<string, []>("input_121_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2306_cast = softmax(axis = var_10, x = input_121_cast)[name = tensor<string, []>("op_2306_cast")];
            tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_49_cast = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = var_2306_cast, y = value_states_51_cast)[name = tensor<string, []>("attn_output_49_cast")];
            tensor<int32, [4]> var_2309_perm_0 = const()[name = tensor<string, []>("op_2309_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2311 = const()[name = tensor<string, []>("op_2311"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_45 = transpose(perm = var_2309_perm_0, x = attn_output_49_cast)[name = tensor<string, []>("transpose_45")];
            tensor<fp16, [1, 128, 2048]> input_123_cast = reshape(shape = var_2311, x = transpose_45)[name = tensor<string, []>("input_123_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_12_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299779328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301876544))), name = tensor<string, []>("model_model_layers_12_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_87_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_12_self_attn_o_proj_weight_to_fp16_palettized, x = input_123_cast)[name = tensor<string, []>("linear_87_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_381_cast = add(x = hidden_states_361_cast, y = linear_87_cast)[name = tensor<string, []>("hidden_states_381_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_25 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_25"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2318_cast = pow(x = hidden_states_381_cast, y = var_13_promoted_to_fp16_25)[name = tensor<string, []>("op_2318_cast")];
            tensor<int32, [1]> var_2319 = const()[name = tensor<string, []>("op_2319"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_51_cast = reduce_mean(axes = var_2319, keep_dims = var_23, x = var_2318_cast)[name = tensor<string, []>("variance_51_cast")];
            tensor<fp16, []> var_2321_to_fp16 = const()[name = tensor<string, []>("op_2321_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2322_cast = add(x = variance_51_cast, y = var_2321_to_fp16)[name = tensor<string, []>("op_2322_cast")];
            tensor<fp16, []> var_2323_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2323_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2323_cast = rsqrt(epsilon = var_2323_epsilon_0_to_fp16, x = var_2322_cast)[name = tensor<string, []>("op_2323_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_385_cast = mul(x = hidden_states_381_cast, y = var_2323_cast)[name = tensor<string, []>("hidden_states_385_cast")];
            tensor<fp16, [2048]> model_model_layers_12_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_12_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301876672)))];
            tensor<fp16, [1, 128, 2048]> input_125_cast = mul(x = model_model_layers_12_post_attention_layernorm_weight_to_fp16, y = hidden_states_385_cast)[name = tensor<string, []>("input_125_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_12_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301880832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307648064))), name = tensor<string, []>("model_model_layers_12_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_88_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_12_mlp_gate_proj_weight_to_fp16_palettized, x = input_125_cast)[name = tensor<string, []>("linear_88_cast")];
            tensor<fp16, [1, 128, 5632]> var_2335_cast = silu(x = linear_88_cast)[name = tensor<string, []>("op_2335_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_12_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307648192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(313415424))), name = tensor<string, []>("model_model_layers_12_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_89_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_12_mlp_up_proj_weight_to_fp16_palettized, x = input_125_cast)[name = tensor<string, []>("linear_89_cast")];
            tensor<fp16, [1, 128, 5632]> input_129_cast = mul(x = var_2335_cast, y = linear_89_cast)[name = tensor<string, []>("input_129_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_12_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(313415552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(319182784))), name = tensor<string, []>("model_model_layers_12_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_90_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_12_mlp_down_proj_weight_to_fp16_palettized, x = input_129_cast)[name = tensor<string, []>("linear_90_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_391_cast = add(x = hidden_states_381_cast, y = linear_90_cast)[name = tensor<string, []>("hidden_states_391_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_26 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_26"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2348_cast = pow(x = hidden_states_391_cast, y = var_13_promoted_to_fp16_26)[name = tensor<string, []>("op_2348_cast")];
            tensor<int32, [1]> var_2349 = const()[name = tensor<string, []>("op_2349"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_53_cast = reduce_mean(axes = var_2349, keep_dims = var_23, x = var_2348_cast)[name = tensor<string, []>("variance_53_cast")];
            tensor<fp16, []> var_2351_to_fp16 = const()[name = tensor<string, []>("op_2351_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2352_cast = add(x = variance_53_cast, y = var_2351_to_fp16)[name = tensor<string, []>("op_2352_cast")];
            tensor<fp16, []> var_2353_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2353_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2353_cast = rsqrt(epsilon = var_2353_epsilon_0_to_fp16, x = var_2352_cast)[name = tensor<string, []>("op_2353_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_395_cast = mul(x = hidden_states_391_cast, y = var_2353_cast)[name = tensor<string, []>("hidden_states_395_cast")];
            tensor<fp16, [2048]> model_model_layers_13_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_13_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(319182912)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_399_cast = mul(x = model_model_layers_13_input_layernorm_weight_to_fp16, y = hidden_states_395_cast)[name = tensor<string, []>("hidden_states_399_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_13_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(319187072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321284288))), name = tensor<string, []>("model_model_layers_13_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_91_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_13_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_399_cast)[name = tensor<string, []>("linear_91_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_13_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321284416))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321546624))), name = tensor<string, []>("model_model_layers_13_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_92_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_13_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_399_cast)[name = tensor<string, []>("linear_92_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_13_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321546752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321808960))), name = tensor<string, []>("model_model_layers_13_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_93_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_13_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_399_cast)[name = tensor<string, []>("linear_93_cast")];
            tensor<int32, [4]> var_2373 = const()[name = tensor<string, []>("op_2373"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2374_cast = reshape(shape = var_2373, x = linear_91_cast)[name = tensor<string, []>("op_2374_cast")];
            tensor<int32, [4]> q_27_perm_0 = const()[name = tensor<string, []>("q_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2376 = const()[name = tensor<string, []>("op_2376"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2377_cast = reshape(shape = var_2376, x = linear_92_cast)[name = tensor<string, []>("op_2377_cast")];
            tensor<int32, [4]> key_states_81_perm_0 = const()[name = tensor<string, []>("key_states_81_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2379 = const()[name = tensor<string, []>("op_2379"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2380_cast = reshape(shape = var_2379, x = linear_93_cast)[name = tensor<string, []>("op_2380_cast")];
            tensor<int32, [4]> hidden_states_405_perm_0 = const()[name = tensor<string, []>("hidden_states_405_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_44 = transpose(perm = q_27_perm_0, x = var_2374_cast)[name = tensor<string, []>("transpose_44")];
            tensor<fp16, [1, 32, 128, 64]> var_2406_cast = mul(x = transpose_44, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2406_cast")];
            tensor<int32, [4]> x1_53_begin_0 = const()[name = tensor<string, []>("x1_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_53_end_0 = const()[name = tensor<string, []>("x1_53_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_53_end_mask_0 = const()[name = tensor<string, []>("x1_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_53_cast = slice_by_index(begin = x1_53_begin_0, end = x1_53_end_0, end_mask = x1_53_end_mask_0, x = transpose_44)[name = tensor<string, []>("x1_53_cast")];
            tensor<int32, [4]> x2_53_begin_0 = const()[name = tensor<string, []>("x2_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_53_end_0 = const()[name = tensor<string, []>("x2_53_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_53_end_mask_0 = const()[name = tensor<string, []>("x2_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_53_cast = slice_by_index(begin = x2_53_begin_0, end = x2_53_end_0, end_mask = x2_53_end_mask_0, x = transpose_44)[name = tensor<string, []>("x2_53_cast")];
            tensor<fp16, []> const_232_promoted_to_fp16 = const()[name = tensor<string, []>("const_232_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2417_cast = mul(x = x2_53_cast, y = const_232_promoted_to_fp16)[name = tensor<string, []>("op_2417_cast")];
            tensor<bool, []> var_2419_interleave_0 = const()[name = tensor<string, []>("op_2419_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2419_cast = concat(axis = var_10, interleave = var_2419_interleave_0, values = (var_2417_cast, x1_53_cast))[name = tensor<string, []>("op_2419_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2420_cast = mul(x = var_2419_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2420_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_55_cast = add(x = var_2406_cast, y = var_2420_cast)[name = tensor<string, []>("query_states_55_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_43 = transpose(perm = key_states_81_perm_0, x = var_2377_cast)[name = tensor<string, []>("transpose_43")];
            tensor<fp16, [1, 4, 128, 64]> var_2422_cast = mul(x = transpose_43, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2422_cast")];
            tensor<int32, [4]> x1_55_begin_0 = const()[name = tensor<string, []>("x1_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_55_end_0 = const()[name = tensor<string, []>("x1_55_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_55_end_mask_0 = const()[name = tensor<string, []>("x1_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_55_cast = slice_by_index(begin = x1_55_begin_0, end = x1_55_end_0, end_mask = x1_55_end_mask_0, x = transpose_43)[name = tensor<string, []>("x1_55_cast")];
            tensor<int32, [4]> x2_55_begin_0 = const()[name = tensor<string, []>("x2_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_55_end_0 = const()[name = tensor<string, []>("x2_55_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_55_end_mask_0 = const()[name = tensor<string, []>("x2_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_55_cast = slice_by_index(begin = x2_55_begin_0, end = x2_55_end_0, end_mask = x2_55_end_mask_0, x = transpose_43)[name = tensor<string, []>("x2_55_cast")];
            tensor<fp16, []> const_235_promoted_to_fp16 = const()[name = tensor<string, []>("const_235_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2433_cast = mul(x = x2_55_cast, y = const_235_promoted_to_fp16)[name = tensor<string, []>("op_2433_cast")];
            tensor<bool, []> var_2435_interleave_0 = const()[name = tensor<string, []>("op_2435_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2435_cast = concat(axis = var_10, interleave = var_2435_interleave_0, values = (var_2433_cast, x1_55_cast))[name = tensor<string, []>("op_2435_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2436_cast = mul(x = var_2435_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2436_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_401_cast = add(x = var_2422_cast, y = var_2436_cast)[name = tensor<string, []>("hidden_states_401_cast")];
            tensor<int32, [1]> var_2445_axes_0 = const()[name = tensor<string, []>("op_2445_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2445_cast = expand_dims(axes = var_2445_axes_0, x = hidden_states_401_cast)[name = tensor<string, []>("op_2445_cast")];
            tensor<int32, [5]> hidden_states_403_reps_0 = const()[name = tensor<string, []>("hidden_states_403_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_403_cast = tile(reps = hidden_states_403_reps_0, x = var_2445_cast)[name = tensor<string, []>("hidden_states_403_cast")];
            tensor<int32, [4]> var_2453 = const()[name = tensor<string, []>("op_2453"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_83_cast = reshape(shape = var_2453, x = hidden_states_403_cast)[name = tensor<string, []>("key_states_83_cast")];
            tensor<int32, [1]> var_2462_axes_0 = const()[name = tensor<string, []>("op_2462_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_42 = transpose(perm = hidden_states_405_perm_0, x = var_2380_cast)[name = tensor<string, []>("transpose_42")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2462_cast = expand_dims(axes = var_2462_axes_0, x = transpose_42)[name = tensor<string, []>("op_2462_cast")];
            tensor<int32, [5]> hidden_states_407_reps_0 = const()[name = tensor<string, []>("hidden_states_407_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_407_cast = tile(reps = hidden_states_407_reps_0, x = var_2462_cast)[name = tensor<string, []>("hidden_states_407_cast")];
            tensor<int32, [4]> var_2470 = const()[name = tensor<string, []>("op_2470"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_55_cast = reshape(shape = var_2470, x = hidden_states_407_cast)[name = tensor<string, []>("value_states_55_cast")];
            tensor<int32, [4]> var_2472_perm_0 = const()[name = tensor<string, []>("op_2472_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2473_transpose_x_0 = const()[name = tensor<string, []>("op_2473_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2473_transpose_y_0 = const()[name = tensor<string, []>("op_2473_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_41 = transpose(perm = var_2472_perm_0, x = key_states_83_cast)[name = tensor<string, []>("transpose_41")];
            tensor<fp16, [1, 32, 128, 128]> var_2473_cast = matmul(transpose_x = var_2473_transpose_x_0, transpose_y = var_2473_transpose_y_0, x = query_states_55_cast, y = transpose_41)[name = tensor<string, []>("op_2473_cast")];
            tensor<fp16, []> _inversed_attn_weights_53_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_53_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_53_cast = mul(x = var_2473_cast, y = _inversed_attn_weights_53_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_53_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_131_cast = add(x = _inversed_attn_weights_53_cast, y = attention_mask_cast)[name = tensor<string, []>("input_131_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2477_cast = softmax(axis = var_10, x = input_131_cast)[name = tensor<string, []>("op_2477_cast")];
            tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_53_cast = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = var_2477_cast, y = value_states_55_cast)[name = tensor<string, []>("attn_output_53_cast")];
            tensor<int32, [4]> var_2480_perm_0 = const()[name = tensor<string, []>("op_2480_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2482 = const()[name = tensor<string, []>("op_2482"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_40 = transpose(perm = var_2480_perm_0, x = attn_output_53_cast)[name = tensor<string, []>("transpose_40")];
            tensor<fp16, [1, 128, 2048]> input_133_cast = reshape(shape = var_2482, x = transpose_40)[name = tensor<string, []>("input_133_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_13_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(321809088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323906304))), name = tensor<string, []>("model_model_layers_13_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_94_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_13_self_attn_o_proj_weight_to_fp16_palettized, x = input_133_cast)[name = tensor<string, []>("linear_94_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_411_cast = add(x = hidden_states_391_cast, y = linear_94_cast)[name = tensor<string, []>("hidden_states_411_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_27 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_27"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2489_cast = pow(x = hidden_states_411_cast, y = var_13_promoted_to_fp16_27)[name = tensor<string, []>("op_2489_cast")];
            tensor<int32, [1]> var_2490 = const()[name = tensor<string, []>("op_2490"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_55_cast = reduce_mean(axes = var_2490, keep_dims = var_23, x = var_2489_cast)[name = tensor<string, []>("variance_55_cast")];
            tensor<fp16, []> var_2492_to_fp16 = const()[name = tensor<string, []>("op_2492_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2493_cast = add(x = variance_55_cast, y = var_2492_to_fp16)[name = tensor<string, []>("op_2493_cast")];
            tensor<fp16, []> var_2494_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2494_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2494_cast = rsqrt(epsilon = var_2494_epsilon_0_to_fp16, x = var_2493_cast)[name = tensor<string, []>("op_2494_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_415_cast = mul(x = hidden_states_411_cast, y = var_2494_cast)[name = tensor<string, []>("hidden_states_415_cast")];
            tensor<fp16, [2048]> model_model_layers_13_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_13_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323906432)))];
            tensor<fp16, [1, 128, 2048]> input_135_cast = mul(x = model_model_layers_13_post_attention_layernorm_weight_to_fp16, y = hidden_states_415_cast)[name = tensor<string, []>("input_135_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_13_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(323910592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329677824))), name = tensor<string, []>("model_model_layers_13_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_95_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_13_mlp_gate_proj_weight_to_fp16_palettized, x = input_135_cast)[name = tensor<string, []>("linear_95_cast")];
            tensor<fp16, [1, 128, 5632]> var_2506_cast = silu(x = linear_95_cast)[name = tensor<string, []>("op_2506_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_13_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329677952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335445184))), name = tensor<string, []>("model_model_layers_13_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_96_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_13_mlp_up_proj_weight_to_fp16_palettized, x = input_135_cast)[name = tensor<string, []>("linear_96_cast")];
            tensor<fp16, [1, 128, 5632]> input_139_cast = mul(x = var_2506_cast, y = linear_96_cast)[name = tensor<string, []>("input_139_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_13_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335445312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341212544))), name = tensor<string, []>("model_model_layers_13_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_97_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_13_mlp_down_proj_weight_to_fp16_palettized, x = input_139_cast)[name = tensor<string, []>("linear_97_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_421_cast = add(x = hidden_states_411_cast, y = linear_97_cast)[name = tensor<string, []>("hidden_states_421_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_28 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_28"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2519_cast = pow(x = hidden_states_421_cast, y = var_13_promoted_to_fp16_28)[name = tensor<string, []>("op_2519_cast")];
            tensor<int32, [1]> var_2520 = const()[name = tensor<string, []>("op_2520"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_57_cast = reduce_mean(axes = var_2520, keep_dims = var_23, x = var_2519_cast)[name = tensor<string, []>("variance_57_cast")];
            tensor<fp16, []> var_2522_to_fp16 = const()[name = tensor<string, []>("op_2522_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2523_cast = add(x = variance_57_cast, y = var_2522_to_fp16)[name = tensor<string, []>("op_2523_cast")];
            tensor<fp16, []> var_2524_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2524_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2524_cast = rsqrt(epsilon = var_2524_epsilon_0_to_fp16, x = var_2523_cast)[name = tensor<string, []>("op_2524_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_425_cast = mul(x = hidden_states_421_cast, y = var_2524_cast)[name = tensor<string, []>("hidden_states_425_cast")];
            tensor<fp16, [2048]> model_model_layers_14_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_14_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341212672)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_429_cast = mul(x = model_model_layers_14_input_layernorm_weight_to_fp16, y = hidden_states_425_cast)[name = tensor<string, []>("hidden_states_429_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_14_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341216832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343314048))), name = tensor<string, []>("model_model_layers_14_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_98_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_14_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_429_cast)[name = tensor<string, []>("linear_98_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_14_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343314176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343576384))), name = tensor<string, []>("model_model_layers_14_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_99_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_14_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_429_cast)[name = tensor<string, []>("linear_99_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_14_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343576512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343838720))), name = tensor<string, []>("model_model_layers_14_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_100_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_14_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_429_cast)[name = tensor<string, []>("linear_100_cast")];
            tensor<int32, [4]> var_2544 = const()[name = tensor<string, []>("op_2544"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2545_cast = reshape(shape = var_2544, x = linear_98_cast)[name = tensor<string, []>("op_2545_cast")];
            tensor<int32, [4]> q_29_perm_0 = const()[name = tensor<string, []>("q_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2547 = const()[name = tensor<string, []>("op_2547"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2548_cast = reshape(shape = var_2547, x = linear_99_cast)[name = tensor<string, []>("op_2548_cast")];
            tensor<int32, [4]> key_states_87_perm_0 = const()[name = tensor<string, []>("key_states_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2550 = const()[name = tensor<string, []>("op_2550"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2551_cast = reshape(shape = var_2550, x = linear_100_cast)[name = tensor<string, []>("op_2551_cast")];
            tensor<int32, [4]> hidden_states_435_perm_0 = const()[name = tensor<string, []>("hidden_states_435_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_39 = transpose(perm = q_29_perm_0, x = var_2545_cast)[name = tensor<string, []>("transpose_39")];
            tensor<fp16, [1, 32, 128, 64]> var_2577_cast = mul(x = transpose_39, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2577_cast")];
            tensor<int32, [4]> x1_57_begin_0 = const()[name = tensor<string, []>("x1_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_57_end_0 = const()[name = tensor<string, []>("x1_57_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_57_end_mask_0 = const()[name = tensor<string, []>("x1_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_57_cast = slice_by_index(begin = x1_57_begin_0, end = x1_57_end_0, end_mask = x1_57_end_mask_0, x = transpose_39)[name = tensor<string, []>("x1_57_cast")];
            tensor<int32, [4]> x2_57_begin_0 = const()[name = tensor<string, []>("x2_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_57_end_0 = const()[name = tensor<string, []>("x2_57_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_57_end_mask_0 = const()[name = tensor<string, []>("x2_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_57_cast = slice_by_index(begin = x2_57_begin_0, end = x2_57_end_0, end_mask = x2_57_end_mask_0, x = transpose_39)[name = tensor<string, []>("x2_57_cast")];
            tensor<fp16, []> const_249_promoted_to_fp16 = const()[name = tensor<string, []>("const_249_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2588_cast = mul(x = x2_57_cast, y = const_249_promoted_to_fp16)[name = tensor<string, []>("op_2588_cast")];
            tensor<bool, []> var_2590_interleave_0 = const()[name = tensor<string, []>("op_2590_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2590_cast = concat(axis = var_10, interleave = var_2590_interleave_0, values = (var_2588_cast, x1_57_cast))[name = tensor<string, []>("op_2590_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2591_cast = mul(x = var_2590_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2591_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_59_cast = add(x = var_2577_cast, y = var_2591_cast)[name = tensor<string, []>("query_states_59_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_38 = transpose(perm = key_states_87_perm_0, x = var_2548_cast)[name = tensor<string, []>("transpose_38")];
            tensor<fp16, [1, 4, 128, 64]> var_2593_cast = mul(x = transpose_38, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2593_cast")];
            tensor<int32, [4]> x1_59_begin_0 = const()[name = tensor<string, []>("x1_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_59_end_0 = const()[name = tensor<string, []>("x1_59_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_59_end_mask_0 = const()[name = tensor<string, []>("x1_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_59_cast = slice_by_index(begin = x1_59_begin_0, end = x1_59_end_0, end_mask = x1_59_end_mask_0, x = transpose_38)[name = tensor<string, []>("x1_59_cast")];
            tensor<int32, [4]> x2_59_begin_0 = const()[name = tensor<string, []>("x2_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_59_end_0 = const()[name = tensor<string, []>("x2_59_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_59_end_mask_0 = const()[name = tensor<string, []>("x2_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_59_cast = slice_by_index(begin = x2_59_begin_0, end = x2_59_end_0, end_mask = x2_59_end_mask_0, x = transpose_38)[name = tensor<string, []>("x2_59_cast")];
            tensor<fp16, []> const_252_promoted_to_fp16 = const()[name = tensor<string, []>("const_252_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2604_cast = mul(x = x2_59_cast, y = const_252_promoted_to_fp16)[name = tensor<string, []>("op_2604_cast")];
            tensor<bool, []> var_2606_interleave_0 = const()[name = tensor<string, []>("op_2606_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2606_cast = concat(axis = var_10, interleave = var_2606_interleave_0, values = (var_2604_cast, x1_59_cast))[name = tensor<string, []>("op_2606_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2607_cast = mul(x = var_2606_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2607_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_431_cast = add(x = var_2593_cast, y = var_2607_cast)[name = tensor<string, []>("hidden_states_431_cast")];
            tensor<int32, [1]> var_2616_axes_0 = const()[name = tensor<string, []>("op_2616_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2616_cast = expand_dims(axes = var_2616_axes_0, x = hidden_states_431_cast)[name = tensor<string, []>("op_2616_cast")];
            tensor<int32, [5]> hidden_states_433_reps_0 = const()[name = tensor<string, []>("hidden_states_433_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_433_cast = tile(reps = hidden_states_433_reps_0, x = var_2616_cast)[name = tensor<string, []>("hidden_states_433_cast")];
            tensor<int32, [4]> var_2624 = const()[name = tensor<string, []>("op_2624"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_89_cast = reshape(shape = var_2624, x = hidden_states_433_cast)[name = tensor<string, []>("key_states_89_cast")];
            tensor<int32, [1]> var_2633_axes_0 = const()[name = tensor<string, []>("op_2633_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_37 = transpose(perm = hidden_states_435_perm_0, x = var_2551_cast)[name = tensor<string, []>("transpose_37")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2633_cast = expand_dims(axes = var_2633_axes_0, x = transpose_37)[name = tensor<string, []>("op_2633_cast")];
            tensor<int32, [5]> hidden_states_437_reps_0 = const()[name = tensor<string, []>("hidden_states_437_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_437_cast = tile(reps = hidden_states_437_reps_0, x = var_2633_cast)[name = tensor<string, []>("hidden_states_437_cast")];
            tensor<int32, [4]> var_2641 = const()[name = tensor<string, []>("op_2641"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_59_cast = reshape(shape = var_2641, x = hidden_states_437_cast)[name = tensor<string, []>("value_states_59_cast")];
            tensor<int32, [4]> var_2643_perm_0 = const()[name = tensor<string, []>("op_2643_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2644_transpose_x_0 = const()[name = tensor<string, []>("op_2644_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2644_transpose_y_0 = const()[name = tensor<string, []>("op_2644_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_36 = transpose(perm = var_2643_perm_0, x = key_states_89_cast)[name = tensor<string, []>("transpose_36")];
            tensor<fp16, [1, 32, 128, 128]> var_2644_cast = matmul(transpose_x = var_2644_transpose_x_0, transpose_y = var_2644_transpose_y_0, x = query_states_59_cast, y = transpose_36)[name = tensor<string, []>("op_2644_cast")];
            tensor<fp16, []> _inversed_attn_weights_57_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_57_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_57_cast = mul(x = var_2644_cast, y = _inversed_attn_weights_57_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_57_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_141_cast = add(x = _inversed_attn_weights_57_cast, y = attention_mask_cast)[name = tensor<string, []>("input_141_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2648_cast = softmax(axis = var_10, x = input_141_cast)[name = tensor<string, []>("op_2648_cast")];
            tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_57_cast = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = var_2648_cast, y = value_states_59_cast)[name = tensor<string, []>("attn_output_57_cast")];
            tensor<int32, [4]> var_2651_perm_0 = const()[name = tensor<string, []>("op_2651_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2653 = const()[name = tensor<string, []>("op_2653"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_35 = transpose(perm = var_2651_perm_0, x = attn_output_57_cast)[name = tensor<string, []>("transpose_35")];
            tensor<fp16, [1, 128, 2048]> input_143_cast = reshape(shape = var_2653, x = transpose_35)[name = tensor<string, []>("input_143_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_14_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343838848))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345936064))), name = tensor<string, []>("model_model_layers_14_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_101_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_14_self_attn_o_proj_weight_to_fp16_palettized, x = input_143_cast)[name = tensor<string, []>("linear_101_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_441_cast = add(x = hidden_states_421_cast, y = linear_101_cast)[name = tensor<string, []>("hidden_states_441_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_29 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_29"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2660_cast = pow(x = hidden_states_441_cast, y = var_13_promoted_to_fp16_29)[name = tensor<string, []>("op_2660_cast")];
            tensor<int32, [1]> var_2661 = const()[name = tensor<string, []>("op_2661"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_59_cast = reduce_mean(axes = var_2661, keep_dims = var_23, x = var_2660_cast)[name = tensor<string, []>("variance_59_cast")];
            tensor<fp16, []> var_2663_to_fp16 = const()[name = tensor<string, []>("op_2663_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2664_cast = add(x = variance_59_cast, y = var_2663_to_fp16)[name = tensor<string, []>("op_2664_cast")];
            tensor<fp16, []> var_2665_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2665_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2665_cast = rsqrt(epsilon = var_2665_epsilon_0_to_fp16, x = var_2664_cast)[name = tensor<string, []>("op_2665_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_445_cast = mul(x = hidden_states_441_cast, y = var_2665_cast)[name = tensor<string, []>("hidden_states_445_cast")];
            tensor<fp16, [2048]> model_model_layers_14_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_14_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345936192)))];
            tensor<fp16, [1, 128, 2048]> input_145_cast = mul(x = model_model_layers_14_post_attention_layernorm_weight_to_fp16, y = hidden_states_445_cast)[name = tensor<string, []>("input_145_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_14_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345940352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351707584))), name = tensor<string, []>("model_model_layers_14_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_102_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_14_mlp_gate_proj_weight_to_fp16_palettized, x = input_145_cast)[name = tensor<string, []>("linear_102_cast")];
            tensor<fp16, [1, 128, 5632]> var_2677_cast = silu(x = linear_102_cast)[name = tensor<string, []>("op_2677_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_14_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(351707712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357474944))), name = tensor<string, []>("model_model_layers_14_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_103_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_14_mlp_up_proj_weight_to_fp16_palettized, x = input_145_cast)[name = tensor<string, []>("linear_103_cast")];
            tensor<fp16, [1, 128, 5632]> input_149_cast = mul(x = var_2677_cast, y = linear_103_cast)[name = tensor<string, []>("input_149_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_14_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(357475072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363242304))), name = tensor<string, []>("model_model_layers_14_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_104_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_14_mlp_down_proj_weight_to_fp16_palettized, x = input_149_cast)[name = tensor<string, []>("linear_104_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_451_cast = add(x = hidden_states_441_cast, y = linear_104_cast)[name = tensor<string, []>("hidden_states_451_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_30 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_30"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2690_cast = pow(x = hidden_states_451_cast, y = var_13_promoted_to_fp16_30)[name = tensor<string, []>("op_2690_cast")];
            tensor<int32, [1]> var_2691 = const()[name = tensor<string, []>("op_2691"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_61_cast = reduce_mean(axes = var_2691, keep_dims = var_23, x = var_2690_cast)[name = tensor<string, []>("variance_61_cast")];
            tensor<fp16, []> var_2693_to_fp16 = const()[name = tensor<string, []>("op_2693_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2694_cast = add(x = variance_61_cast, y = var_2693_to_fp16)[name = tensor<string, []>("op_2694_cast")];
            tensor<fp16, []> var_2695_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2695_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2695_cast = rsqrt(epsilon = var_2695_epsilon_0_to_fp16, x = var_2694_cast)[name = tensor<string, []>("op_2695_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_455_cast = mul(x = hidden_states_451_cast, y = var_2695_cast)[name = tensor<string, []>("hidden_states_455_cast")];
            tensor<fp16, [2048]> model_model_layers_15_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_15_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363242432)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_459_cast = mul(x = model_model_layers_15_input_layernorm_weight_to_fp16, y = hidden_states_455_cast)[name = tensor<string, []>("hidden_states_459_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_15_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(363246592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365343808))), name = tensor<string, []>("model_model_layers_15_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_105_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_15_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_459_cast)[name = tensor<string, []>("linear_105_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_15_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365343936))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365606144))), name = tensor<string, []>("model_model_layers_15_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_106_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_15_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_459_cast)[name = tensor<string, []>("linear_106_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_15_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365606272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365868480))), name = tensor<string, []>("model_model_layers_15_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_107_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_15_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_459_cast)[name = tensor<string, []>("linear_107_cast")];
            tensor<int32, [4]> var_2715 = const()[name = tensor<string, []>("op_2715"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2716_cast = reshape(shape = var_2715, x = linear_105_cast)[name = tensor<string, []>("op_2716_cast")];
            tensor<int32, [4]> q_31_perm_0 = const()[name = tensor<string, []>("q_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2718 = const()[name = tensor<string, []>("op_2718"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2719_cast = reshape(shape = var_2718, x = linear_106_cast)[name = tensor<string, []>("op_2719_cast")];
            tensor<int32, [4]> key_states_93_perm_0 = const()[name = tensor<string, []>("key_states_93_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2721 = const()[name = tensor<string, []>("op_2721"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2722_cast = reshape(shape = var_2721, x = linear_107_cast)[name = tensor<string, []>("op_2722_cast")];
            tensor<int32, [4]> hidden_states_465_perm_0 = const()[name = tensor<string, []>("hidden_states_465_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_34 = transpose(perm = q_31_perm_0, x = var_2716_cast)[name = tensor<string, []>("transpose_34")];
            tensor<fp16, [1, 32, 128, 64]> var_2748_cast = mul(x = transpose_34, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2748_cast")];
            tensor<int32, [4]> x1_61_begin_0 = const()[name = tensor<string, []>("x1_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_61_end_0 = const()[name = tensor<string, []>("x1_61_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_61_end_mask_0 = const()[name = tensor<string, []>("x1_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_61_cast = slice_by_index(begin = x1_61_begin_0, end = x1_61_end_0, end_mask = x1_61_end_mask_0, x = transpose_34)[name = tensor<string, []>("x1_61_cast")];
            tensor<int32, [4]> x2_61_begin_0 = const()[name = tensor<string, []>("x2_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_61_end_0 = const()[name = tensor<string, []>("x2_61_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_61_end_mask_0 = const()[name = tensor<string, []>("x2_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_61_cast = slice_by_index(begin = x2_61_begin_0, end = x2_61_end_0, end_mask = x2_61_end_mask_0, x = transpose_34)[name = tensor<string, []>("x2_61_cast")];
            tensor<fp16, []> const_266_promoted_to_fp16 = const()[name = tensor<string, []>("const_266_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2759_cast = mul(x = x2_61_cast, y = const_266_promoted_to_fp16)[name = tensor<string, []>("op_2759_cast")];
            tensor<bool, []> var_2761_interleave_0 = const()[name = tensor<string, []>("op_2761_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2761_cast = concat(axis = var_10, interleave = var_2761_interleave_0, values = (var_2759_cast, x1_61_cast))[name = tensor<string, []>("op_2761_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2762_cast = mul(x = var_2761_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2762_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_63_cast = add(x = var_2748_cast, y = var_2762_cast)[name = tensor<string, []>("query_states_63_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_33 = transpose(perm = key_states_93_perm_0, x = var_2719_cast)[name = tensor<string, []>("transpose_33")];
            tensor<fp16, [1, 4, 128, 64]> var_2764_cast = mul(x = transpose_33, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2764_cast")];
            tensor<int32, [4]> x1_63_begin_0 = const()[name = tensor<string, []>("x1_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_63_end_0 = const()[name = tensor<string, []>("x1_63_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_63_end_mask_0 = const()[name = tensor<string, []>("x1_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_63_cast = slice_by_index(begin = x1_63_begin_0, end = x1_63_end_0, end_mask = x1_63_end_mask_0, x = transpose_33)[name = tensor<string, []>("x1_63_cast")];
            tensor<int32, [4]> x2_63_begin_0 = const()[name = tensor<string, []>("x2_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_63_end_0 = const()[name = tensor<string, []>("x2_63_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_63_end_mask_0 = const()[name = tensor<string, []>("x2_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_63_cast = slice_by_index(begin = x2_63_begin_0, end = x2_63_end_0, end_mask = x2_63_end_mask_0, x = transpose_33)[name = tensor<string, []>("x2_63_cast")];
            tensor<fp16, []> const_269_promoted_to_fp16 = const()[name = tensor<string, []>("const_269_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2775_cast = mul(x = x2_63_cast, y = const_269_promoted_to_fp16)[name = tensor<string, []>("op_2775_cast")];
            tensor<bool, []> var_2777_interleave_0 = const()[name = tensor<string, []>("op_2777_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2777_cast = concat(axis = var_10, interleave = var_2777_interleave_0, values = (var_2775_cast, x1_63_cast))[name = tensor<string, []>("op_2777_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2778_cast = mul(x = var_2777_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2778_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_461_cast = add(x = var_2764_cast, y = var_2778_cast)[name = tensor<string, []>("hidden_states_461_cast")];
            tensor<int32, [1]> var_2787_axes_0 = const()[name = tensor<string, []>("op_2787_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2787_cast = expand_dims(axes = var_2787_axes_0, x = hidden_states_461_cast)[name = tensor<string, []>("op_2787_cast")];
            tensor<int32, [5]> hidden_states_463_reps_0 = const()[name = tensor<string, []>("hidden_states_463_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_463_cast = tile(reps = hidden_states_463_reps_0, x = var_2787_cast)[name = tensor<string, []>("hidden_states_463_cast")];
            tensor<int32, [4]> var_2795 = const()[name = tensor<string, []>("op_2795"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_95_cast = reshape(shape = var_2795, x = hidden_states_463_cast)[name = tensor<string, []>("key_states_95_cast")];
            tensor<int32, [1]> var_2804_axes_0 = const()[name = tensor<string, []>("op_2804_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_32 = transpose(perm = hidden_states_465_perm_0, x = var_2722_cast)[name = tensor<string, []>("transpose_32")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2804_cast = expand_dims(axes = var_2804_axes_0, x = transpose_32)[name = tensor<string, []>("op_2804_cast")];
            tensor<int32, [5]> hidden_states_467_reps_0 = const()[name = tensor<string, []>("hidden_states_467_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_467_cast = tile(reps = hidden_states_467_reps_0, x = var_2804_cast)[name = tensor<string, []>("hidden_states_467_cast")];
            tensor<int32, [4]> var_2812 = const()[name = tensor<string, []>("op_2812"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_63_cast = reshape(shape = var_2812, x = hidden_states_467_cast)[name = tensor<string, []>("value_states_63_cast")];
            tensor<int32, [4]> var_2814_perm_0 = const()[name = tensor<string, []>("op_2814_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2815_transpose_x_0 = const()[name = tensor<string, []>("op_2815_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2815_transpose_y_0 = const()[name = tensor<string, []>("op_2815_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_31 = transpose(perm = var_2814_perm_0, x = key_states_95_cast)[name = tensor<string, []>("transpose_31")];
            tensor<fp16, [1, 32, 128, 128]> var_2815_cast = matmul(transpose_x = var_2815_transpose_x_0, transpose_y = var_2815_transpose_y_0, x = query_states_63_cast, y = transpose_31)[name = tensor<string, []>("op_2815_cast")];
            tensor<fp16, []> _inversed_attn_weights_61_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_61_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_61_cast = mul(x = var_2815_cast, y = _inversed_attn_weights_61_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_61_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_151_cast = add(x = _inversed_attn_weights_61_cast, y = attention_mask_cast)[name = tensor<string, []>("input_151_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2819_cast = softmax(axis = var_10, x = input_151_cast)[name = tensor<string, []>("op_2819_cast")];
            tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_61_cast = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = var_2819_cast, y = value_states_63_cast)[name = tensor<string, []>("attn_output_61_cast")];
            tensor<int32, [4]> var_2822_perm_0 = const()[name = tensor<string, []>("op_2822_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2824 = const()[name = tensor<string, []>("op_2824"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_30 = transpose(perm = var_2822_perm_0, x = attn_output_61_cast)[name = tensor<string, []>("transpose_30")];
            tensor<fp16, [1, 128, 2048]> input_153_cast = reshape(shape = var_2824, x = transpose_30)[name = tensor<string, []>("input_153_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_15_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365868608))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367965824))), name = tensor<string, []>("model_model_layers_15_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_108_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_15_self_attn_o_proj_weight_to_fp16_palettized, x = input_153_cast)[name = tensor<string, []>("linear_108_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_471_cast = add(x = hidden_states_451_cast, y = linear_108_cast)[name = tensor<string, []>("hidden_states_471_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_31 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_31"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2831_cast = pow(x = hidden_states_471_cast, y = var_13_promoted_to_fp16_31)[name = tensor<string, []>("op_2831_cast")];
            tensor<int32, [1]> var_2832 = const()[name = tensor<string, []>("op_2832"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_63_cast = reduce_mean(axes = var_2832, keep_dims = var_23, x = var_2831_cast)[name = tensor<string, []>("variance_63_cast")];
            tensor<fp16, []> var_2834_to_fp16 = const()[name = tensor<string, []>("op_2834_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2835_cast = add(x = variance_63_cast, y = var_2834_to_fp16)[name = tensor<string, []>("op_2835_cast")];
            tensor<fp16, []> var_2836_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2836_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2836_cast = rsqrt(epsilon = var_2836_epsilon_0_to_fp16, x = var_2835_cast)[name = tensor<string, []>("op_2836_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_475_cast = mul(x = hidden_states_471_cast, y = var_2836_cast)[name = tensor<string, []>("hidden_states_475_cast")];
            tensor<fp16, [2048]> model_model_layers_15_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_15_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367965952)))];
            tensor<fp16, [1, 128, 2048]> input_155_cast = mul(x = model_model_layers_15_post_attention_layernorm_weight_to_fp16, y = hidden_states_475_cast)[name = tensor<string, []>("input_155_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_15_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367970112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373737344))), name = tensor<string, []>("model_model_layers_15_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_109_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_15_mlp_gate_proj_weight_to_fp16_palettized, x = input_155_cast)[name = tensor<string, []>("linear_109_cast")];
            tensor<fp16, [1, 128, 5632]> var_2848_cast = silu(x = linear_109_cast)[name = tensor<string, []>("op_2848_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_15_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373737472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379504704))), name = tensor<string, []>("model_model_layers_15_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_110_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_15_mlp_up_proj_weight_to_fp16_palettized, x = input_155_cast)[name = tensor<string, []>("linear_110_cast")];
            tensor<fp16, [1, 128, 5632]> input_159_cast = mul(x = var_2848_cast, y = linear_110_cast)[name = tensor<string, []>("input_159_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_15_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(379504832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385272064))), name = tensor<string, []>("model_model_layers_15_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_111_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_15_mlp_down_proj_weight_to_fp16_palettized, x = input_159_cast)[name = tensor<string, []>("linear_111_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_481_cast = add(x = hidden_states_471_cast, y = linear_111_cast)[name = tensor<string, []>("hidden_states_481_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_32 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_32"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_2861_cast = pow(x = hidden_states_481_cast, y = var_13_promoted_to_fp16_32)[name = tensor<string, []>("op_2861_cast")];
            tensor<int32, [1]> var_2862 = const()[name = tensor<string, []>("op_2862"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_65_cast = reduce_mean(axes = var_2862, keep_dims = var_23, x = var_2861_cast)[name = tensor<string, []>("variance_65_cast")];
            tensor<fp16, []> var_2864_to_fp16 = const()[name = tensor<string, []>("op_2864_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_2865_cast = add(x = variance_65_cast, y = var_2864_to_fp16)[name = tensor<string, []>("op_2865_cast")];
            tensor<fp16, []> var_2866_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_2866_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_2866_cast = rsqrt(epsilon = var_2866_epsilon_0_to_fp16, x = var_2865_cast)[name = tensor<string, []>("op_2866_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_485_cast = mul(x = hidden_states_481_cast, y = var_2866_cast)[name = tensor<string, []>("hidden_states_485_cast")];
            tensor<fp16, [2048]> model_model_layers_16_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_16_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385272192)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_489_cast = mul(x = model_model_layers_16_input_layernorm_weight_to_fp16, y = hidden_states_485_cast)[name = tensor<string, []>("hidden_states_489_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_16_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(385276352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387373568))), name = tensor<string, []>("model_model_layers_16_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_112_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_16_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_489_cast)[name = tensor<string, []>("linear_112_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_16_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387373696))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387635904))), name = tensor<string, []>("model_model_layers_16_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_113_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_16_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_489_cast)[name = tensor<string, []>("linear_113_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_16_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387636032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387898240))), name = tensor<string, []>("model_model_layers_16_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_114_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_16_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_489_cast)[name = tensor<string, []>("linear_114_cast")];
            tensor<int32, [4]> var_2886 = const()[name = tensor<string, []>("op_2886"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_2887_cast = reshape(shape = var_2886, x = linear_112_cast)[name = tensor<string, []>("op_2887_cast")];
            tensor<int32, [4]> q_33_perm_0 = const()[name = tensor<string, []>("q_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2889 = const()[name = tensor<string, []>("op_2889"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2890_cast = reshape(shape = var_2889, x = linear_113_cast)[name = tensor<string, []>("op_2890_cast")];
            tensor<int32, [4]> key_states_99_perm_0 = const()[name = tensor<string, []>("key_states_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_2892 = const()[name = tensor<string, []>("op_2892"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_2893_cast = reshape(shape = var_2892, x = linear_114_cast)[name = tensor<string, []>("op_2893_cast")];
            tensor<int32, [4]> hidden_states_495_perm_0 = const()[name = tensor<string, []>("hidden_states_495_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_29 = transpose(perm = q_33_perm_0, x = var_2887_cast)[name = tensor<string, []>("transpose_29")];
            tensor<fp16, [1, 32, 128, 64]> var_2919_cast = mul(x = transpose_29, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2919_cast")];
            tensor<int32, [4]> x1_65_begin_0 = const()[name = tensor<string, []>("x1_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_65_end_0 = const()[name = tensor<string, []>("x1_65_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_65_end_mask_0 = const()[name = tensor<string, []>("x1_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_65_cast = slice_by_index(begin = x1_65_begin_0, end = x1_65_end_0, end_mask = x1_65_end_mask_0, x = transpose_29)[name = tensor<string, []>("x1_65_cast")];
            tensor<int32, [4]> x2_65_begin_0 = const()[name = tensor<string, []>("x2_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_65_end_0 = const()[name = tensor<string, []>("x2_65_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_65_end_mask_0 = const()[name = tensor<string, []>("x2_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_65_cast = slice_by_index(begin = x2_65_begin_0, end = x2_65_end_0, end_mask = x2_65_end_mask_0, x = transpose_29)[name = tensor<string, []>("x2_65_cast")];
            tensor<fp16, []> const_283_promoted_to_fp16 = const()[name = tensor<string, []>("const_283_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_2930_cast = mul(x = x2_65_cast, y = const_283_promoted_to_fp16)[name = tensor<string, []>("op_2930_cast")];
            tensor<bool, []> var_2932_interleave_0 = const()[name = tensor<string, []>("op_2932_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_2932_cast = concat(axis = var_10, interleave = var_2932_interleave_0, values = (var_2930_cast, x1_65_cast))[name = tensor<string, []>("op_2932_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_2933_cast = mul(x = var_2932_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2933_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_67_cast = add(x = var_2919_cast, y = var_2933_cast)[name = tensor<string, []>("query_states_67_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_28 = transpose(perm = key_states_99_perm_0, x = var_2890_cast)[name = tensor<string, []>("transpose_28")];
            tensor<fp16, [1, 4, 128, 64]> var_2935_cast = mul(x = transpose_28, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_2935_cast")];
            tensor<int32, [4]> x1_67_begin_0 = const()[name = tensor<string, []>("x1_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_67_end_0 = const()[name = tensor<string, []>("x1_67_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_67_end_mask_0 = const()[name = tensor<string, []>("x1_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_67_cast = slice_by_index(begin = x1_67_begin_0, end = x1_67_end_0, end_mask = x1_67_end_mask_0, x = transpose_28)[name = tensor<string, []>("x1_67_cast")];
            tensor<int32, [4]> x2_67_begin_0 = const()[name = tensor<string, []>("x2_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_67_end_0 = const()[name = tensor<string, []>("x2_67_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_67_end_mask_0 = const()[name = tensor<string, []>("x2_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_67_cast = slice_by_index(begin = x2_67_begin_0, end = x2_67_end_0, end_mask = x2_67_end_mask_0, x = transpose_28)[name = tensor<string, []>("x2_67_cast")];
            tensor<fp16, []> const_286_promoted_to_fp16 = const()[name = tensor<string, []>("const_286_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_2946_cast = mul(x = x2_67_cast, y = const_286_promoted_to_fp16)[name = tensor<string, []>("op_2946_cast")];
            tensor<bool, []> var_2948_interleave_0 = const()[name = tensor<string, []>("op_2948_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_2948_cast = concat(axis = var_10, interleave = var_2948_interleave_0, values = (var_2946_cast, x1_67_cast))[name = tensor<string, []>("op_2948_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_2949_cast = mul(x = var_2948_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_2949_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_491_cast = add(x = var_2935_cast, y = var_2949_cast)[name = tensor<string, []>("hidden_states_491_cast")];
            tensor<int32, [1]> var_2958_axes_0 = const()[name = tensor<string, []>("op_2958_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2958_cast = expand_dims(axes = var_2958_axes_0, x = hidden_states_491_cast)[name = tensor<string, []>("op_2958_cast")];
            tensor<int32, [5]> hidden_states_493_reps_0 = const()[name = tensor<string, []>("hidden_states_493_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_493_cast = tile(reps = hidden_states_493_reps_0, x = var_2958_cast)[name = tensor<string, []>("hidden_states_493_cast")];
            tensor<int32, [4]> var_2966 = const()[name = tensor<string, []>("op_2966"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_101_cast = reshape(shape = var_2966, x = hidden_states_493_cast)[name = tensor<string, []>("key_states_101_cast")];
            tensor<int32, [1]> var_2975_axes_0 = const()[name = tensor<string, []>("op_2975_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_27 = transpose(perm = hidden_states_495_perm_0, x = var_2893_cast)[name = tensor<string, []>("transpose_27")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_2975_cast = expand_dims(axes = var_2975_axes_0, x = transpose_27)[name = tensor<string, []>("op_2975_cast")];
            tensor<int32, [5]> hidden_states_497_reps_0 = const()[name = tensor<string, []>("hidden_states_497_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_497_cast = tile(reps = hidden_states_497_reps_0, x = var_2975_cast)[name = tensor<string, []>("hidden_states_497_cast")];
            tensor<int32, [4]> var_2983 = const()[name = tensor<string, []>("op_2983"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_67_cast = reshape(shape = var_2983, x = hidden_states_497_cast)[name = tensor<string, []>("value_states_67_cast")];
            tensor<int32, [4]> var_2985_perm_0 = const()[name = tensor<string, []>("op_2985_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_2986_transpose_x_0 = const()[name = tensor<string, []>("op_2986_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_2986_transpose_y_0 = const()[name = tensor<string, []>("op_2986_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_26 = transpose(perm = var_2985_perm_0, x = key_states_101_cast)[name = tensor<string, []>("transpose_26")];
            tensor<fp16, [1, 32, 128, 128]> var_2986_cast = matmul(transpose_x = var_2986_transpose_x_0, transpose_y = var_2986_transpose_y_0, x = query_states_67_cast, y = transpose_26)[name = tensor<string, []>("op_2986_cast")];
            tensor<fp16, []> _inversed_attn_weights_65_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_65_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_65_cast = mul(x = var_2986_cast, y = _inversed_attn_weights_65_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_65_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_161_cast = add(x = _inversed_attn_weights_65_cast, y = attention_mask_cast)[name = tensor<string, []>("input_161_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_2990_cast = softmax(axis = var_10, x = input_161_cast)[name = tensor<string, []>("op_2990_cast")];
            tensor<bool, []> attn_output_65_transpose_x_0 = const()[name = tensor<string, []>("attn_output_65_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_65_transpose_y_0 = const()[name = tensor<string, []>("attn_output_65_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_65_cast = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = var_2990_cast, y = value_states_67_cast)[name = tensor<string, []>("attn_output_65_cast")];
            tensor<int32, [4]> var_2993_perm_0 = const()[name = tensor<string, []>("op_2993_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2995 = const()[name = tensor<string, []>("op_2995"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_25 = transpose(perm = var_2993_perm_0, x = attn_output_65_cast)[name = tensor<string, []>("transpose_25")];
            tensor<fp16, [1, 128, 2048]> input_163_cast = reshape(shape = var_2995, x = transpose_25)[name = tensor<string, []>("input_163_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_16_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(387898368))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389995584))), name = tensor<string, []>("model_model_layers_16_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_115_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_16_self_attn_o_proj_weight_to_fp16_palettized, x = input_163_cast)[name = tensor<string, []>("linear_115_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_501_cast = add(x = hidden_states_481_cast, y = linear_115_cast)[name = tensor<string, []>("hidden_states_501_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_33 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_33"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3002_cast = pow(x = hidden_states_501_cast, y = var_13_promoted_to_fp16_33)[name = tensor<string, []>("op_3002_cast")];
            tensor<int32, [1]> var_3003 = const()[name = tensor<string, []>("op_3003"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_67_cast = reduce_mean(axes = var_3003, keep_dims = var_23, x = var_3002_cast)[name = tensor<string, []>("variance_67_cast")];
            tensor<fp16, []> var_3005_to_fp16 = const()[name = tensor<string, []>("op_3005_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3006_cast = add(x = variance_67_cast, y = var_3005_to_fp16)[name = tensor<string, []>("op_3006_cast")];
            tensor<fp16, []> var_3007_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3007_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3007_cast = rsqrt(epsilon = var_3007_epsilon_0_to_fp16, x = var_3006_cast)[name = tensor<string, []>("op_3007_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_505_cast = mul(x = hidden_states_501_cast, y = var_3007_cast)[name = tensor<string, []>("hidden_states_505_cast")];
            tensor<fp16, [2048]> model_model_layers_16_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_16_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389995712)))];
            tensor<fp16, [1, 128, 2048]> input_165_cast = mul(x = model_model_layers_16_post_attention_layernorm_weight_to_fp16, y = hidden_states_505_cast)[name = tensor<string, []>("input_165_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_16_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(389999872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395767104))), name = tensor<string, []>("model_model_layers_16_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_116_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_16_mlp_gate_proj_weight_to_fp16_palettized, x = input_165_cast)[name = tensor<string, []>("linear_116_cast")];
            tensor<fp16, [1, 128, 5632]> var_3019_cast = silu(x = linear_116_cast)[name = tensor<string, []>("op_3019_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_16_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(395767232))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401534464))), name = tensor<string, []>("model_model_layers_16_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_117_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_16_mlp_up_proj_weight_to_fp16_palettized, x = input_165_cast)[name = tensor<string, []>("linear_117_cast")];
            tensor<fp16, [1, 128, 5632]> input_169_cast = mul(x = var_3019_cast, y = linear_117_cast)[name = tensor<string, []>("input_169_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_16_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401534592))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407301824))), name = tensor<string, []>("model_model_layers_16_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_118_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_16_mlp_down_proj_weight_to_fp16_palettized, x = input_169_cast)[name = tensor<string, []>("linear_118_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_511_cast = add(x = hidden_states_501_cast, y = linear_118_cast)[name = tensor<string, []>("hidden_states_511_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_34 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_34"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3032_cast = pow(x = hidden_states_511_cast, y = var_13_promoted_to_fp16_34)[name = tensor<string, []>("op_3032_cast")];
            tensor<int32, [1]> var_3033 = const()[name = tensor<string, []>("op_3033"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_69_cast = reduce_mean(axes = var_3033, keep_dims = var_23, x = var_3032_cast)[name = tensor<string, []>("variance_69_cast")];
            tensor<fp16, []> var_3035_to_fp16 = const()[name = tensor<string, []>("op_3035_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3036_cast = add(x = variance_69_cast, y = var_3035_to_fp16)[name = tensor<string, []>("op_3036_cast")];
            tensor<fp16, []> var_3037_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3037_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3037_cast = rsqrt(epsilon = var_3037_epsilon_0_to_fp16, x = var_3036_cast)[name = tensor<string, []>("op_3037_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_515_cast = mul(x = hidden_states_511_cast, y = var_3037_cast)[name = tensor<string, []>("hidden_states_515_cast")];
            tensor<fp16, [2048]> model_model_layers_17_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_17_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407301952)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_519_cast = mul(x = model_model_layers_17_input_layernorm_weight_to_fp16, y = hidden_states_515_cast)[name = tensor<string, []>("hidden_states_519_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_17_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407306112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409403328))), name = tensor<string, []>("model_model_layers_17_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_119_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_17_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_519_cast)[name = tensor<string, []>("linear_119_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_17_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409403456))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409665664))), name = tensor<string, []>("model_model_layers_17_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_120_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_17_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_519_cast)[name = tensor<string, []>("linear_120_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_17_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409665792))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409928000))), name = tensor<string, []>("model_model_layers_17_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_121_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_17_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_519_cast)[name = tensor<string, []>("linear_121_cast")];
            tensor<int32, [4]> var_3057 = const()[name = tensor<string, []>("op_3057"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_3058_cast = reshape(shape = var_3057, x = linear_119_cast)[name = tensor<string, []>("op_3058_cast")];
            tensor<int32, [4]> q_35_perm_0 = const()[name = tensor<string, []>("q_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3060 = const()[name = tensor<string, []>("op_3060"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3061_cast = reshape(shape = var_3060, x = linear_120_cast)[name = tensor<string, []>("op_3061_cast")];
            tensor<int32, [4]> key_states_105_perm_0 = const()[name = tensor<string, []>("key_states_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3063 = const()[name = tensor<string, []>("op_3063"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3064_cast = reshape(shape = var_3063, x = linear_121_cast)[name = tensor<string, []>("op_3064_cast")];
            tensor<int32, [4]> hidden_states_525_perm_0 = const()[name = tensor<string, []>("hidden_states_525_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_24 = transpose(perm = q_35_perm_0, x = var_3058_cast)[name = tensor<string, []>("transpose_24")];
            tensor<fp16, [1, 32, 128, 64]> var_3090_cast = mul(x = transpose_24, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3090_cast")];
            tensor<int32, [4]> x1_69_begin_0 = const()[name = tensor<string, []>("x1_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_69_end_0 = const()[name = tensor<string, []>("x1_69_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_69_end_mask_0 = const()[name = tensor<string, []>("x1_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_69_cast = slice_by_index(begin = x1_69_begin_0, end = x1_69_end_0, end_mask = x1_69_end_mask_0, x = transpose_24)[name = tensor<string, []>("x1_69_cast")];
            tensor<int32, [4]> x2_69_begin_0 = const()[name = tensor<string, []>("x2_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_69_end_0 = const()[name = tensor<string, []>("x2_69_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_69_end_mask_0 = const()[name = tensor<string, []>("x2_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_69_cast = slice_by_index(begin = x2_69_begin_0, end = x2_69_end_0, end_mask = x2_69_end_mask_0, x = transpose_24)[name = tensor<string, []>("x2_69_cast")];
            tensor<fp16, []> const_300_promoted_to_fp16 = const()[name = tensor<string, []>("const_300_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_3101_cast = mul(x = x2_69_cast, y = const_300_promoted_to_fp16)[name = tensor<string, []>("op_3101_cast")];
            tensor<bool, []> var_3103_interleave_0 = const()[name = tensor<string, []>("op_3103_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_3103_cast = concat(axis = var_10, interleave = var_3103_interleave_0, values = (var_3101_cast, x1_69_cast))[name = tensor<string, []>("op_3103_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_3104_cast = mul(x = var_3103_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3104_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_71_cast = add(x = var_3090_cast, y = var_3104_cast)[name = tensor<string, []>("query_states_71_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_23 = transpose(perm = key_states_105_perm_0, x = var_3061_cast)[name = tensor<string, []>("transpose_23")];
            tensor<fp16, [1, 4, 128, 64]> var_3106_cast = mul(x = transpose_23, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3106_cast")];
            tensor<int32, [4]> x1_71_begin_0 = const()[name = tensor<string, []>("x1_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_71_end_0 = const()[name = tensor<string, []>("x1_71_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_71_end_mask_0 = const()[name = tensor<string, []>("x1_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_71_cast = slice_by_index(begin = x1_71_begin_0, end = x1_71_end_0, end_mask = x1_71_end_mask_0, x = transpose_23)[name = tensor<string, []>("x1_71_cast")];
            tensor<int32, [4]> x2_71_begin_0 = const()[name = tensor<string, []>("x2_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_71_end_0 = const()[name = tensor<string, []>("x2_71_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_71_end_mask_0 = const()[name = tensor<string, []>("x2_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_71_cast = slice_by_index(begin = x2_71_begin_0, end = x2_71_end_0, end_mask = x2_71_end_mask_0, x = transpose_23)[name = tensor<string, []>("x2_71_cast")];
            tensor<fp16, []> const_303_promoted_to_fp16 = const()[name = tensor<string, []>("const_303_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_3117_cast = mul(x = x2_71_cast, y = const_303_promoted_to_fp16)[name = tensor<string, []>("op_3117_cast")];
            tensor<bool, []> var_3119_interleave_0 = const()[name = tensor<string, []>("op_3119_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_3119_cast = concat(axis = var_10, interleave = var_3119_interleave_0, values = (var_3117_cast, x1_71_cast))[name = tensor<string, []>("op_3119_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_3120_cast = mul(x = var_3119_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3120_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_521_cast = add(x = var_3106_cast, y = var_3120_cast)[name = tensor<string, []>("hidden_states_521_cast")];
            tensor<int32, [1]> var_3129_axes_0 = const()[name = tensor<string, []>("op_3129_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3129_cast = expand_dims(axes = var_3129_axes_0, x = hidden_states_521_cast)[name = tensor<string, []>("op_3129_cast")];
            tensor<int32, [5]> hidden_states_523_reps_0 = const()[name = tensor<string, []>("hidden_states_523_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_523_cast = tile(reps = hidden_states_523_reps_0, x = var_3129_cast)[name = tensor<string, []>("hidden_states_523_cast")];
            tensor<int32, [4]> var_3137 = const()[name = tensor<string, []>("op_3137"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_107_cast = reshape(shape = var_3137, x = hidden_states_523_cast)[name = tensor<string, []>("key_states_107_cast")];
            tensor<int32, [1]> var_3146_axes_0 = const()[name = tensor<string, []>("op_3146_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_22 = transpose(perm = hidden_states_525_perm_0, x = var_3064_cast)[name = tensor<string, []>("transpose_22")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3146_cast = expand_dims(axes = var_3146_axes_0, x = transpose_22)[name = tensor<string, []>("op_3146_cast")];
            tensor<int32, [5]> hidden_states_527_reps_0 = const()[name = tensor<string, []>("hidden_states_527_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_527_cast = tile(reps = hidden_states_527_reps_0, x = var_3146_cast)[name = tensor<string, []>("hidden_states_527_cast")];
            tensor<int32, [4]> var_3154 = const()[name = tensor<string, []>("op_3154"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_71_cast = reshape(shape = var_3154, x = hidden_states_527_cast)[name = tensor<string, []>("value_states_71_cast")];
            tensor<int32, [4]> var_3156_perm_0 = const()[name = tensor<string, []>("op_3156_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_3157_transpose_x_0 = const()[name = tensor<string, []>("op_3157_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3157_transpose_y_0 = const()[name = tensor<string, []>("op_3157_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_21 = transpose(perm = var_3156_perm_0, x = key_states_107_cast)[name = tensor<string, []>("transpose_21")];
            tensor<fp16, [1, 32, 128, 128]> var_3157_cast = matmul(transpose_x = var_3157_transpose_x_0, transpose_y = var_3157_transpose_y_0, x = query_states_71_cast, y = transpose_21)[name = tensor<string, []>("op_3157_cast")];
            tensor<fp16, []> _inversed_attn_weights_69_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_69_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_69_cast = mul(x = var_3157_cast, y = _inversed_attn_weights_69_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_69_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_171_cast = add(x = _inversed_attn_weights_69_cast, y = attention_mask_cast)[name = tensor<string, []>("input_171_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_3161_cast = softmax(axis = var_10, x = input_171_cast)[name = tensor<string, []>("op_3161_cast")];
            tensor<bool, []> attn_output_69_transpose_x_0 = const()[name = tensor<string, []>("attn_output_69_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_69_transpose_y_0 = const()[name = tensor<string, []>("attn_output_69_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_69_cast = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = var_3161_cast, y = value_states_71_cast)[name = tensor<string, []>("attn_output_69_cast")];
            tensor<int32, [4]> var_3164_perm_0 = const()[name = tensor<string, []>("op_3164_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3166 = const()[name = tensor<string, []>("op_3166"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_20 = transpose(perm = var_3164_perm_0, x = attn_output_69_cast)[name = tensor<string, []>("transpose_20")];
            tensor<fp16, [1, 128, 2048]> input_173_cast = reshape(shape = var_3166, x = transpose_20)[name = tensor<string, []>("input_173_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_17_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(409928128))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412025344))), name = tensor<string, []>("model_model_layers_17_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_122_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_17_self_attn_o_proj_weight_to_fp16_palettized, x = input_173_cast)[name = tensor<string, []>("linear_122_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_531_cast = add(x = hidden_states_511_cast, y = linear_122_cast)[name = tensor<string, []>("hidden_states_531_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_35 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_35"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3173_cast = pow(x = hidden_states_531_cast, y = var_13_promoted_to_fp16_35)[name = tensor<string, []>("op_3173_cast")];
            tensor<int32, [1]> var_3174 = const()[name = tensor<string, []>("op_3174"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_71_cast = reduce_mean(axes = var_3174, keep_dims = var_23, x = var_3173_cast)[name = tensor<string, []>("variance_71_cast")];
            tensor<fp16, []> var_3176_to_fp16 = const()[name = tensor<string, []>("op_3176_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3177_cast = add(x = variance_71_cast, y = var_3176_to_fp16)[name = tensor<string, []>("op_3177_cast")];
            tensor<fp16, []> var_3178_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3178_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3178_cast = rsqrt(epsilon = var_3178_epsilon_0_to_fp16, x = var_3177_cast)[name = tensor<string, []>("op_3178_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_535_cast = mul(x = hidden_states_531_cast, y = var_3178_cast)[name = tensor<string, []>("hidden_states_535_cast")];
            tensor<fp16, [2048]> model_model_layers_17_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_17_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412025472)))];
            tensor<fp16, [1, 128, 2048]> input_175_cast = mul(x = model_model_layers_17_post_attention_layernorm_weight_to_fp16, y = hidden_states_535_cast)[name = tensor<string, []>("input_175_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_17_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(412029632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417796864))), name = tensor<string, []>("model_model_layers_17_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_123_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_17_mlp_gate_proj_weight_to_fp16_palettized, x = input_175_cast)[name = tensor<string, []>("linear_123_cast")];
            tensor<fp16, [1, 128, 5632]> var_3190_cast = silu(x = linear_123_cast)[name = tensor<string, []>("op_3190_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_17_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417796992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(423564224))), name = tensor<string, []>("model_model_layers_17_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_124_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_17_mlp_up_proj_weight_to_fp16_palettized, x = input_175_cast)[name = tensor<string, []>("linear_124_cast")];
            tensor<fp16, [1, 128, 5632]> input_179_cast = mul(x = var_3190_cast, y = linear_124_cast)[name = tensor<string, []>("input_179_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_17_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(423564352))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429331584))), name = tensor<string, []>("model_model_layers_17_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_125_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_17_mlp_down_proj_weight_to_fp16_palettized, x = input_179_cast)[name = tensor<string, []>("linear_125_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_541_cast = add(x = hidden_states_531_cast, y = linear_125_cast)[name = tensor<string, []>("hidden_states_541_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_36 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_36"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3203_cast = pow(x = hidden_states_541_cast, y = var_13_promoted_to_fp16_36)[name = tensor<string, []>("op_3203_cast")];
            tensor<int32, [1]> var_3204 = const()[name = tensor<string, []>("op_3204"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_73_cast = reduce_mean(axes = var_3204, keep_dims = var_23, x = var_3203_cast)[name = tensor<string, []>("variance_73_cast")];
            tensor<fp16, []> var_3206_to_fp16 = const()[name = tensor<string, []>("op_3206_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3207_cast = add(x = variance_73_cast, y = var_3206_to_fp16)[name = tensor<string, []>("op_3207_cast")];
            tensor<fp16, []> var_3208_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3208_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3208_cast = rsqrt(epsilon = var_3208_epsilon_0_to_fp16, x = var_3207_cast)[name = tensor<string, []>("op_3208_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_545_cast = mul(x = hidden_states_541_cast, y = var_3208_cast)[name = tensor<string, []>("hidden_states_545_cast")];
            tensor<fp16, [2048]> model_model_layers_18_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_18_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429331712)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_549_cast = mul(x = model_model_layers_18_input_layernorm_weight_to_fp16, y = hidden_states_545_cast)[name = tensor<string, []>("hidden_states_549_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_18_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429335872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431433088))), name = tensor<string, []>("model_model_layers_18_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_126_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_18_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_549_cast)[name = tensor<string, []>("linear_126_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_18_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431433216))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431695424))), name = tensor<string, []>("model_model_layers_18_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_127_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_18_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_549_cast)[name = tensor<string, []>("linear_127_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_18_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431695552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431957760))), name = tensor<string, []>("model_model_layers_18_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_128_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_18_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_549_cast)[name = tensor<string, []>("linear_128_cast")];
            tensor<int32, [4]> var_3228 = const()[name = tensor<string, []>("op_3228"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_3229_cast = reshape(shape = var_3228, x = linear_126_cast)[name = tensor<string, []>("op_3229_cast")];
            tensor<int32, [4]> q_37_perm_0 = const()[name = tensor<string, []>("q_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3231 = const()[name = tensor<string, []>("op_3231"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3232_cast = reshape(shape = var_3231, x = linear_127_cast)[name = tensor<string, []>("op_3232_cast")];
            tensor<int32, [4]> key_states_111_perm_0 = const()[name = tensor<string, []>("key_states_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3234 = const()[name = tensor<string, []>("op_3234"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3235_cast = reshape(shape = var_3234, x = linear_128_cast)[name = tensor<string, []>("op_3235_cast")];
            tensor<int32, [4]> hidden_states_555_perm_0 = const()[name = tensor<string, []>("hidden_states_555_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_19 = transpose(perm = q_37_perm_0, x = var_3229_cast)[name = tensor<string, []>("transpose_19")];
            tensor<fp16, [1, 32, 128, 64]> var_3261_cast = mul(x = transpose_19, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3261_cast")];
            tensor<int32, [4]> x1_73_begin_0 = const()[name = tensor<string, []>("x1_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_73_end_0 = const()[name = tensor<string, []>("x1_73_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_73_end_mask_0 = const()[name = tensor<string, []>("x1_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_73_cast = slice_by_index(begin = x1_73_begin_0, end = x1_73_end_0, end_mask = x1_73_end_mask_0, x = transpose_19)[name = tensor<string, []>("x1_73_cast")];
            tensor<int32, [4]> x2_73_begin_0 = const()[name = tensor<string, []>("x2_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_73_end_0 = const()[name = tensor<string, []>("x2_73_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_73_end_mask_0 = const()[name = tensor<string, []>("x2_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_73_cast = slice_by_index(begin = x2_73_begin_0, end = x2_73_end_0, end_mask = x2_73_end_mask_0, x = transpose_19)[name = tensor<string, []>("x2_73_cast")];
            tensor<fp16, []> const_317_promoted_to_fp16 = const()[name = tensor<string, []>("const_317_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_3272_cast = mul(x = x2_73_cast, y = const_317_promoted_to_fp16)[name = tensor<string, []>("op_3272_cast")];
            tensor<bool, []> var_3274_interleave_0 = const()[name = tensor<string, []>("op_3274_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_3274_cast = concat(axis = var_10, interleave = var_3274_interleave_0, values = (var_3272_cast, x1_73_cast))[name = tensor<string, []>("op_3274_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_3275_cast = mul(x = var_3274_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3275_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_75_cast = add(x = var_3261_cast, y = var_3275_cast)[name = tensor<string, []>("query_states_75_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_18 = transpose(perm = key_states_111_perm_0, x = var_3232_cast)[name = tensor<string, []>("transpose_18")];
            tensor<fp16, [1, 4, 128, 64]> var_3277_cast = mul(x = transpose_18, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3277_cast")];
            tensor<int32, [4]> x1_75_begin_0 = const()[name = tensor<string, []>("x1_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_75_end_0 = const()[name = tensor<string, []>("x1_75_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_75_end_mask_0 = const()[name = tensor<string, []>("x1_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_75_cast = slice_by_index(begin = x1_75_begin_0, end = x1_75_end_0, end_mask = x1_75_end_mask_0, x = transpose_18)[name = tensor<string, []>("x1_75_cast")];
            tensor<int32, [4]> x2_75_begin_0 = const()[name = tensor<string, []>("x2_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_75_end_0 = const()[name = tensor<string, []>("x2_75_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_75_end_mask_0 = const()[name = tensor<string, []>("x2_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_75_cast = slice_by_index(begin = x2_75_begin_0, end = x2_75_end_0, end_mask = x2_75_end_mask_0, x = transpose_18)[name = tensor<string, []>("x2_75_cast")];
            tensor<fp16, []> const_320_promoted_to_fp16 = const()[name = tensor<string, []>("const_320_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_3288_cast = mul(x = x2_75_cast, y = const_320_promoted_to_fp16)[name = tensor<string, []>("op_3288_cast")];
            tensor<bool, []> var_3290_interleave_0 = const()[name = tensor<string, []>("op_3290_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_3290_cast = concat(axis = var_10, interleave = var_3290_interleave_0, values = (var_3288_cast, x1_75_cast))[name = tensor<string, []>("op_3290_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_3291_cast = mul(x = var_3290_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3291_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_551_cast = add(x = var_3277_cast, y = var_3291_cast)[name = tensor<string, []>("hidden_states_551_cast")];
            tensor<int32, [1]> var_3300_axes_0 = const()[name = tensor<string, []>("op_3300_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3300_cast = expand_dims(axes = var_3300_axes_0, x = hidden_states_551_cast)[name = tensor<string, []>("op_3300_cast")];
            tensor<int32, [5]> hidden_states_553_reps_0 = const()[name = tensor<string, []>("hidden_states_553_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_553_cast = tile(reps = hidden_states_553_reps_0, x = var_3300_cast)[name = tensor<string, []>("hidden_states_553_cast")];
            tensor<int32, [4]> var_3308 = const()[name = tensor<string, []>("op_3308"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_113_cast = reshape(shape = var_3308, x = hidden_states_553_cast)[name = tensor<string, []>("key_states_113_cast")];
            tensor<int32, [1]> var_3317_axes_0 = const()[name = tensor<string, []>("op_3317_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_17 = transpose(perm = hidden_states_555_perm_0, x = var_3235_cast)[name = tensor<string, []>("transpose_17")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3317_cast = expand_dims(axes = var_3317_axes_0, x = transpose_17)[name = tensor<string, []>("op_3317_cast")];
            tensor<int32, [5]> hidden_states_557_reps_0 = const()[name = tensor<string, []>("hidden_states_557_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_557_cast = tile(reps = hidden_states_557_reps_0, x = var_3317_cast)[name = tensor<string, []>("hidden_states_557_cast")];
            tensor<int32, [4]> var_3325 = const()[name = tensor<string, []>("op_3325"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_75_cast = reshape(shape = var_3325, x = hidden_states_557_cast)[name = tensor<string, []>("value_states_75_cast")];
            tensor<int32, [4]> var_3327_perm_0 = const()[name = tensor<string, []>("op_3327_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_3328_transpose_x_0 = const()[name = tensor<string, []>("op_3328_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3328_transpose_y_0 = const()[name = tensor<string, []>("op_3328_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_16 = transpose(perm = var_3327_perm_0, x = key_states_113_cast)[name = tensor<string, []>("transpose_16")];
            tensor<fp16, [1, 32, 128, 128]> var_3328_cast = matmul(transpose_x = var_3328_transpose_x_0, transpose_y = var_3328_transpose_y_0, x = query_states_75_cast, y = transpose_16)[name = tensor<string, []>("op_3328_cast")];
            tensor<fp16, []> _inversed_attn_weights_73_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_73_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_73_cast = mul(x = var_3328_cast, y = _inversed_attn_weights_73_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_73_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_181_cast = add(x = _inversed_attn_weights_73_cast, y = attention_mask_cast)[name = tensor<string, []>("input_181_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_3332_cast = softmax(axis = var_10, x = input_181_cast)[name = tensor<string, []>("op_3332_cast")];
            tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_73_cast = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = var_3332_cast, y = value_states_75_cast)[name = tensor<string, []>("attn_output_73_cast")];
            tensor<int32, [4]> var_3335_perm_0 = const()[name = tensor<string, []>("op_3335_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3337 = const()[name = tensor<string, []>("op_3337"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_15 = transpose(perm = var_3335_perm_0, x = attn_output_73_cast)[name = tensor<string, []>("transpose_15")];
            tensor<fp16, [1, 128, 2048]> input_183_cast = reshape(shape = var_3337, x = transpose_15)[name = tensor<string, []>("input_183_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_18_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431957888))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434055104))), name = tensor<string, []>("model_model_layers_18_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_129_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_18_self_attn_o_proj_weight_to_fp16_palettized, x = input_183_cast)[name = tensor<string, []>("linear_129_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_561_cast = add(x = hidden_states_541_cast, y = linear_129_cast)[name = tensor<string, []>("hidden_states_561_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_37 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_37"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3344_cast = pow(x = hidden_states_561_cast, y = var_13_promoted_to_fp16_37)[name = tensor<string, []>("op_3344_cast")];
            tensor<int32, [1]> var_3345 = const()[name = tensor<string, []>("op_3345"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_75_cast = reduce_mean(axes = var_3345, keep_dims = var_23, x = var_3344_cast)[name = tensor<string, []>("variance_75_cast")];
            tensor<fp16, []> var_3347_to_fp16 = const()[name = tensor<string, []>("op_3347_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3348_cast = add(x = variance_75_cast, y = var_3347_to_fp16)[name = tensor<string, []>("op_3348_cast")];
            tensor<fp16, []> var_3349_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3349_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3349_cast = rsqrt(epsilon = var_3349_epsilon_0_to_fp16, x = var_3348_cast)[name = tensor<string, []>("op_3349_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_565_cast = mul(x = hidden_states_561_cast, y = var_3349_cast)[name = tensor<string, []>("hidden_states_565_cast")];
            tensor<fp16, [2048]> model_model_layers_18_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_18_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434055232)))];
            tensor<fp16, [1, 128, 2048]> input_185_cast = mul(x = model_model_layers_18_post_attention_layernorm_weight_to_fp16, y = hidden_states_565_cast)[name = tensor<string, []>("input_185_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_18_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434059392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439826624))), name = tensor<string, []>("model_model_layers_18_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_130_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_18_mlp_gate_proj_weight_to_fp16_palettized, x = input_185_cast)[name = tensor<string, []>("linear_130_cast")];
            tensor<fp16, [1, 128, 5632]> var_3361_cast = silu(x = linear_130_cast)[name = tensor<string, []>("op_3361_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_18_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439826752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445593984))), name = tensor<string, []>("model_model_layers_18_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_131_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_18_mlp_up_proj_weight_to_fp16_palettized, x = input_185_cast)[name = tensor<string, []>("linear_131_cast")];
            tensor<fp16, [1, 128, 5632]> input_189_cast = mul(x = var_3361_cast, y = linear_131_cast)[name = tensor<string, []>("input_189_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_18_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(445594112))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451361344))), name = tensor<string, []>("model_model_layers_18_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_132_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_18_mlp_down_proj_weight_to_fp16_palettized, x = input_189_cast)[name = tensor<string, []>("linear_132_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_571_cast = add(x = hidden_states_561_cast, y = linear_132_cast)[name = tensor<string, []>("hidden_states_571_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_38 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_38"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3374_cast = pow(x = hidden_states_571_cast, y = var_13_promoted_to_fp16_38)[name = tensor<string, []>("op_3374_cast")];
            tensor<int32, [1]> var_3375 = const()[name = tensor<string, []>("op_3375"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_77_cast = reduce_mean(axes = var_3375, keep_dims = var_23, x = var_3374_cast)[name = tensor<string, []>("variance_77_cast")];
            tensor<fp16, []> var_3377_to_fp16 = const()[name = tensor<string, []>("op_3377_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3378_cast = add(x = variance_77_cast, y = var_3377_to_fp16)[name = tensor<string, []>("op_3378_cast")];
            tensor<fp16, []> var_3379_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3379_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3379_cast = rsqrt(epsilon = var_3379_epsilon_0_to_fp16, x = var_3378_cast)[name = tensor<string, []>("op_3379_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_575_cast = mul(x = hidden_states_571_cast, y = var_3379_cast)[name = tensor<string, []>("hidden_states_575_cast")];
            tensor<fp16, [2048]> model_model_layers_19_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_19_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451361472)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_579_cast = mul(x = model_model_layers_19_input_layernorm_weight_to_fp16, y = hidden_states_575_cast)[name = tensor<string, []>("hidden_states_579_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_19_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(451365632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453462848))), name = tensor<string, []>("model_model_layers_19_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_133_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_19_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_579_cast)[name = tensor<string, []>("linear_133_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_19_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453462976))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453725184))), name = tensor<string, []>("model_model_layers_19_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_134_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_19_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_579_cast)[name = tensor<string, []>("linear_134_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_19_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453725312))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453987520))), name = tensor<string, []>("model_model_layers_19_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_135_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_19_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_579_cast)[name = tensor<string, []>("linear_135_cast")];
            tensor<int32, [4]> var_3399 = const()[name = tensor<string, []>("op_3399"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_3400_cast = reshape(shape = var_3399, x = linear_133_cast)[name = tensor<string, []>("op_3400_cast")];
            tensor<int32, [4]> q_39_perm_0 = const()[name = tensor<string, []>("q_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3402 = const()[name = tensor<string, []>("op_3402"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3403_cast = reshape(shape = var_3402, x = linear_134_cast)[name = tensor<string, []>("op_3403_cast")];
            tensor<int32, [4]> key_states_117_perm_0 = const()[name = tensor<string, []>("key_states_117_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3405 = const()[name = tensor<string, []>("op_3405"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3406_cast = reshape(shape = var_3405, x = linear_135_cast)[name = tensor<string, []>("op_3406_cast")];
            tensor<int32, [4]> hidden_states_585_perm_0 = const()[name = tensor<string, []>("hidden_states_585_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_14 = transpose(perm = q_39_perm_0, x = var_3400_cast)[name = tensor<string, []>("transpose_14")];
            tensor<fp16, [1, 32, 128, 64]> var_3432_cast = mul(x = transpose_14, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3432_cast")];
            tensor<int32, [4]> x1_77_begin_0 = const()[name = tensor<string, []>("x1_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_77_end_0 = const()[name = tensor<string, []>("x1_77_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_77_end_mask_0 = const()[name = tensor<string, []>("x1_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_77_cast = slice_by_index(begin = x1_77_begin_0, end = x1_77_end_0, end_mask = x1_77_end_mask_0, x = transpose_14)[name = tensor<string, []>("x1_77_cast")];
            tensor<int32, [4]> x2_77_begin_0 = const()[name = tensor<string, []>("x2_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_77_end_0 = const()[name = tensor<string, []>("x2_77_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_77_end_mask_0 = const()[name = tensor<string, []>("x2_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_77_cast = slice_by_index(begin = x2_77_begin_0, end = x2_77_end_0, end_mask = x2_77_end_mask_0, x = transpose_14)[name = tensor<string, []>("x2_77_cast")];
            tensor<fp16, []> const_334_promoted_to_fp16 = const()[name = tensor<string, []>("const_334_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_3443_cast = mul(x = x2_77_cast, y = const_334_promoted_to_fp16)[name = tensor<string, []>("op_3443_cast")];
            tensor<bool, []> var_3445_interleave_0 = const()[name = tensor<string, []>("op_3445_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_3445_cast = concat(axis = var_10, interleave = var_3445_interleave_0, values = (var_3443_cast, x1_77_cast))[name = tensor<string, []>("op_3445_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_3446_cast = mul(x = var_3445_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3446_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_79_cast = add(x = var_3432_cast, y = var_3446_cast)[name = tensor<string, []>("query_states_79_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_13 = transpose(perm = key_states_117_perm_0, x = var_3403_cast)[name = tensor<string, []>("transpose_13")];
            tensor<fp16, [1, 4, 128, 64]> var_3448_cast = mul(x = transpose_13, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3448_cast")];
            tensor<int32, [4]> x1_79_begin_0 = const()[name = tensor<string, []>("x1_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_79_end_0 = const()[name = tensor<string, []>("x1_79_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_79_end_mask_0 = const()[name = tensor<string, []>("x1_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_79_cast = slice_by_index(begin = x1_79_begin_0, end = x1_79_end_0, end_mask = x1_79_end_mask_0, x = transpose_13)[name = tensor<string, []>("x1_79_cast")];
            tensor<int32, [4]> x2_79_begin_0 = const()[name = tensor<string, []>("x2_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_79_end_0 = const()[name = tensor<string, []>("x2_79_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_79_end_mask_0 = const()[name = tensor<string, []>("x2_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_79_cast = slice_by_index(begin = x2_79_begin_0, end = x2_79_end_0, end_mask = x2_79_end_mask_0, x = transpose_13)[name = tensor<string, []>("x2_79_cast")];
            tensor<fp16, []> const_337_promoted_to_fp16 = const()[name = tensor<string, []>("const_337_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_3459_cast = mul(x = x2_79_cast, y = const_337_promoted_to_fp16)[name = tensor<string, []>("op_3459_cast")];
            tensor<bool, []> var_3461_interleave_0 = const()[name = tensor<string, []>("op_3461_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_3461_cast = concat(axis = var_10, interleave = var_3461_interleave_0, values = (var_3459_cast, x1_79_cast))[name = tensor<string, []>("op_3461_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_3462_cast = mul(x = var_3461_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3462_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_581_cast = add(x = var_3448_cast, y = var_3462_cast)[name = tensor<string, []>("hidden_states_581_cast")];
            tensor<int32, [1]> var_3471_axes_0 = const()[name = tensor<string, []>("op_3471_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3471_cast = expand_dims(axes = var_3471_axes_0, x = hidden_states_581_cast)[name = tensor<string, []>("op_3471_cast")];
            tensor<int32, [5]> hidden_states_583_reps_0 = const()[name = tensor<string, []>("hidden_states_583_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_583_cast = tile(reps = hidden_states_583_reps_0, x = var_3471_cast)[name = tensor<string, []>("hidden_states_583_cast")];
            tensor<int32, [4]> var_3479 = const()[name = tensor<string, []>("op_3479"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_119_cast = reshape(shape = var_3479, x = hidden_states_583_cast)[name = tensor<string, []>("key_states_119_cast")];
            tensor<int32, [1]> var_3488_axes_0 = const()[name = tensor<string, []>("op_3488_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_12 = transpose(perm = hidden_states_585_perm_0, x = var_3406_cast)[name = tensor<string, []>("transpose_12")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3488_cast = expand_dims(axes = var_3488_axes_0, x = transpose_12)[name = tensor<string, []>("op_3488_cast")];
            tensor<int32, [5]> hidden_states_587_reps_0 = const()[name = tensor<string, []>("hidden_states_587_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_587_cast = tile(reps = hidden_states_587_reps_0, x = var_3488_cast)[name = tensor<string, []>("hidden_states_587_cast")];
            tensor<int32, [4]> var_3496 = const()[name = tensor<string, []>("op_3496"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_79_cast = reshape(shape = var_3496, x = hidden_states_587_cast)[name = tensor<string, []>("value_states_79_cast")];
            tensor<int32, [4]> var_3498_perm_0 = const()[name = tensor<string, []>("op_3498_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_3499_transpose_x_0 = const()[name = tensor<string, []>("op_3499_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3499_transpose_y_0 = const()[name = tensor<string, []>("op_3499_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_11 = transpose(perm = var_3498_perm_0, x = key_states_119_cast)[name = tensor<string, []>("transpose_11")];
            tensor<fp16, [1, 32, 128, 128]> var_3499_cast = matmul(transpose_x = var_3499_transpose_x_0, transpose_y = var_3499_transpose_y_0, x = query_states_79_cast, y = transpose_11)[name = tensor<string, []>("op_3499_cast")];
            tensor<fp16, []> _inversed_attn_weights_77_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_77_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_77_cast = mul(x = var_3499_cast, y = _inversed_attn_weights_77_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_77_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_191_cast = add(x = _inversed_attn_weights_77_cast, y = attention_mask_cast)[name = tensor<string, []>("input_191_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_3503_cast = softmax(axis = var_10, x = input_191_cast)[name = tensor<string, []>("op_3503_cast")];
            tensor<bool, []> attn_output_77_transpose_x_0 = const()[name = tensor<string, []>("attn_output_77_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_77_transpose_y_0 = const()[name = tensor<string, []>("attn_output_77_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_77_cast = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = var_3503_cast, y = value_states_79_cast)[name = tensor<string, []>("attn_output_77_cast")];
            tensor<int32, [4]> var_3506_perm_0 = const()[name = tensor<string, []>("op_3506_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3508 = const()[name = tensor<string, []>("op_3508"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_10 = transpose(perm = var_3506_perm_0, x = attn_output_77_cast)[name = tensor<string, []>("transpose_10")];
            tensor<fp16, [1, 128, 2048]> input_193_cast = reshape(shape = var_3508, x = transpose_10)[name = tensor<string, []>("input_193_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_19_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453987648))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(456084864))), name = tensor<string, []>("model_model_layers_19_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_136_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_19_self_attn_o_proj_weight_to_fp16_palettized, x = input_193_cast)[name = tensor<string, []>("linear_136_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_591_cast = add(x = hidden_states_571_cast, y = linear_136_cast)[name = tensor<string, []>("hidden_states_591_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_39 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_39"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3515_cast = pow(x = hidden_states_591_cast, y = var_13_promoted_to_fp16_39)[name = tensor<string, []>("op_3515_cast")];
            tensor<int32, [1]> var_3516 = const()[name = tensor<string, []>("op_3516"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_79_cast = reduce_mean(axes = var_3516, keep_dims = var_23, x = var_3515_cast)[name = tensor<string, []>("variance_79_cast")];
            tensor<fp16, []> var_3518_to_fp16 = const()[name = tensor<string, []>("op_3518_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3519_cast = add(x = variance_79_cast, y = var_3518_to_fp16)[name = tensor<string, []>("op_3519_cast")];
            tensor<fp16, []> var_3520_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3520_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3520_cast = rsqrt(epsilon = var_3520_epsilon_0_to_fp16, x = var_3519_cast)[name = tensor<string, []>("op_3520_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_595_cast = mul(x = hidden_states_591_cast, y = var_3520_cast)[name = tensor<string, []>("hidden_states_595_cast")];
            tensor<fp16, [2048]> model_model_layers_19_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_19_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(456084992)))];
            tensor<fp16, [1, 128, 2048]> input_195_cast = mul(x = model_model_layers_19_post_attention_layernorm_weight_to_fp16, y = hidden_states_595_cast)[name = tensor<string, []>("input_195_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_19_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(456089152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461856384))), name = tensor<string, []>("model_model_layers_19_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_137_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_19_mlp_gate_proj_weight_to_fp16_palettized, x = input_195_cast)[name = tensor<string, []>("linear_137_cast")];
            tensor<fp16, [1, 128, 5632]> var_3532_cast = silu(x = linear_137_cast)[name = tensor<string, []>("op_3532_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_19_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(461856512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467623744))), name = tensor<string, []>("model_model_layers_19_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_138_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_19_mlp_up_proj_weight_to_fp16_palettized, x = input_195_cast)[name = tensor<string, []>("linear_138_cast")];
            tensor<fp16, [1, 128, 5632]> input_199_cast = mul(x = var_3532_cast, y = linear_138_cast)[name = tensor<string, []>("input_199_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_19_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467623872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473391104))), name = tensor<string, []>("model_model_layers_19_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_139_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_19_mlp_down_proj_weight_to_fp16_palettized, x = input_199_cast)[name = tensor<string, []>("linear_139_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_601_cast = add(x = hidden_states_591_cast, y = linear_139_cast)[name = tensor<string, []>("hidden_states_601_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_40 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_40"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3545_cast = pow(x = hidden_states_601_cast, y = var_13_promoted_to_fp16_40)[name = tensor<string, []>("op_3545_cast")];
            tensor<int32, [1]> var_3546 = const()[name = tensor<string, []>("op_3546"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_81_cast = reduce_mean(axes = var_3546, keep_dims = var_23, x = var_3545_cast)[name = tensor<string, []>("variance_81_cast")];
            tensor<fp16, []> var_3548_to_fp16 = const()[name = tensor<string, []>("op_3548_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3549_cast = add(x = variance_81_cast, y = var_3548_to_fp16)[name = tensor<string, []>("op_3549_cast")];
            tensor<fp16, []> var_3550_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3550_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3550_cast = rsqrt(epsilon = var_3550_epsilon_0_to_fp16, x = var_3549_cast)[name = tensor<string, []>("op_3550_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_605_cast = mul(x = hidden_states_601_cast, y = var_3550_cast)[name = tensor<string, []>("hidden_states_605_cast")];
            tensor<fp16, [2048]> model_model_layers_20_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_20_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473391232)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_609_cast = mul(x = model_model_layers_20_input_layernorm_weight_to_fp16, y = hidden_states_605_cast)[name = tensor<string, []>("hidden_states_609_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_20_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473395392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475492608))), name = tensor<string, []>("model_model_layers_20_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_140_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_20_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_609_cast)[name = tensor<string, []>("linear_140_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_20_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475492736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475754944))), name = tensor<string, []>("model_model_layers_20_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_141_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_20_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_609_cast)[name = tensor<string, []>("linear_141_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_20_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(475755072))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476017280))), name = tensor<string, []>("model_model_layers_20_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_142_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_20_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_609_cast)[name = tensor<string, []>("linear_142_cast")];
            tensor<int32, [4]> var_3570 = const()[name = tensor<string, []>("op_3570"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_3571_cast = reshape(shape = var_3570, x = linear_140_cast)[name = tensor<string, []>("op_3571_cast")];
            tensor<int32, [4]> q_41_perm_0 = const()[name = tensor<string, []>("q_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3573 = const()[name = tensor<string, []>("op_3573"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3574_cast = reshape(shape = var_3573, x = linear_141_cast)[name = tensor<string, []>("op_3574_cast")];
            tensor<int32, [4]> key_states_123_perm_0 = const()[name = tensor<string, []>("key_states_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3576 = const()[name = tensor<string, []>("op_3576"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3577_cast = reshape(shape = var_3576, x = linear_142_cast)[name = tensor<string, []>("op_3577_cast")];
            tensor<int32, [4]> hidden_states_615_perm_0 = const()[name = tensor<string, []>("hidden_states_615_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_9 = transpose(perm = q_41_perm_0, x = var_3571_cast)[name = tensor<string, []>("transpose_9")];
            tensor<fp16, [1, 32, 128, 64]> var_3603_cast = mul(x = transpose_9, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3603_cast")];
            tensor<int32, [4]> x1_81_begin_0 = const()[name = tensor<string, []>("x1_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_81_end_0 = const()[name = tensor<string, []>("x1_81_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_81_end_mask_0 = const()[name = tensor<string, []>("x1_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_81_cast = slice_by_index(begin = x1_81_begin_0, end = x1_81_end_0, end_mask = x1_81_end_mask_0, x = transpose_9)[name = tensor<string, []>("x1_81_cast")];
            tensor<int32, [4]> x2_81_begin_0 = const()[name = tensor<string, []>("x2_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_81_end_0 = const()[name = tensor<string, []>("x2_81_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_81_end_mask_0 = const()[name = tensor<string, []>("x2_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_81_cast = slice_by_index(begin = x2_81_begin_0, end = x2_81_end_0, end_mask = x2_81_end_mask_0, x = transpose_9)[name = tensor<string, []>("x2_81_cast")];
            tensor<fp16, []> const_351_promoted_to_fp16 = const()[name = tensor<string, []>("const_351_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_3614_cast = mul(x = x2_81_cast, y = const_351_promoted_to_fp16)[name = tensor<string, []>("op_3614_cast")];
            tensor<bool, []> var_3616_interleave_0 = const()[name = tensor<string, []>("op_3616_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_3616_cast = concat(axis = var_10, interleave = var_3616_interleave_0, values = (var_3614_cast, x1_81_cast))[name = tensor<string, []>("op_3616_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_3617_cast = mul(x = var_3616_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3617_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_83_cast = add(x = var_3603_cast, y = var_3617_cast)[name = tensor<string, []>("query_states_83_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_8 = transpose(perm = key_states_123_perm_0, x = var_3574_cast)[name = tensor<string, []>("transpose_8")];
            tensor<fp16, [1, 4, 128, 64]> var_3619_cast = mul(x = transpose_8, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3619_cast")];
            tensor<int32, [4]> x1_83_begin_0 = const()[name = tensor<string, []>("x1_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_83_end_0 = const()[name = tensor<string, []>("x1_83_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_83_end_mask_0 = const()[name = tensor<string, []>("x1_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_83_cast = slice_by_index(begin = x1_83_begin_0, end = x1_83_end_0, end_mask = x1_83_end_mask_0, x = transpose_8)[name = tensor<string, []>("x1_83_cast")];
            tensor<int32, [4]> x2_83_begin_0 = const()[name = tensor<string, []>("x2_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_83_end_0 = const()[name = tensor<string, []>("x2_83_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_83_end_mask_0 = const()[name = tensor<string, []>("x2_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_83_cast = slice_by_index(begin = x2_83_begin_0, end = x2_83_end_0, end_mask = x2_83_end_mask_0, x = transpose_8)[name = tensor<string, []>("x2_83_cast")];
            tensor<fp16, []> const_354_promoted_to_fp16 = const()[name = tensor<string, []>("const_354_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_3630_cast = mul(x = x2_83_cast, y = const_354_promoted_to_fp16)[name = tensor<string, []>("op_3630_cast")];
            tensor<bool, []> var_3632_interleave_0 = const()[name = tensor<string, []>("op_3632_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_3632_cast = concat(axis = var_10, interleave = var_3632_interleave_0, values = (var_3630_cast, x1_83_cast))[name = tensor<string, []>("op_3632_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_3633_cast = mul(x = var_3632_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3633_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_611_cast = add(x = var_3619_cast, y = var_3633_cast)[name = tensor<string, []>("hidden_states_611_cast")];
            tensor<int32, [1]> var_3642_axes_0 = const()[name = tensor<string, []>("op_3642_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3642_cast = expand_dims(axes = var_3642_axes_0, x = hidden_states_611_cast)[name = tensor<string, []>("op_3642_cast")];
            tensor<int32, [5]> hidden_states_613_reps_0 = const()[name = tensor<string, []>("hidden_states_613_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_613_cast = tile(reps = hidden_states_613_reps_0, x = var_3642_cast)[name = tensor<string, []>("hidden_states_613_cast")];
            tensor<int32, [4]> var_3650 = const()[name = tensor<string, []>("op_3650"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_125_cast = reshape(shape = var_3650, x = hidden_states_613_cast)[name = tensor<string, []>("key_states_125_cast")];
            tensor<int32, [1]> var_3659_axes_0 = const()[name = tensor<string, []>("op_3659_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_7 = transpose(perm = hidden_states_615_perm_0, x = var_3577_cast)[name = tensor<string, []>("transpose_7")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3659_cast = expand_dims(axes = var_3659_axes_0, x = transpose_7)[name = tensor<string, []>("op_3659_cast")];
            tensor<int32, [5]> hidden_states_617_reps_0 = const()[name = tensor<string, []>("hidden_states_617_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_617_cast = tile(reps = hidden_states_617_reps_0, x = var_3659_cast)[name = tensor<string, []>("hidden_states_617_cast")];
            tensor<int32, [4]> var_3667 = const()[name = tensor<string, []>("op_3667"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_83_cast = reshape(shape = var_3667, x = hidden_states_617_cast)[name = tensor<string, []>("value_states_83_cast")];
            tensor<int32, [4]> var_3669_perm_0 = const()[name = tensor<string, []>("op_3669_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_3670_transpose_x_0 = const()[name = tensor<string, []>("op_3670_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3670_transpose_y_0 = const()[name = tensor<string, []>("op_3670_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_6 = transpose(perm = var_3669_perm_0, x = key_states_125_cast)[name = tensor<string, []>("transpose_6")];
            tensor<fp16, [1, 32, 128, 128]> var_3670_cast = matmul(transpose_x = var_3670_transpose_x_0, transpose_y = var_3670_transpose_y_0, x = query_states_83_cast, y = transpose_6)[name = tensor<string, []>("op_3670_cast")];
            tensor<fp16, []> _inversed_attn_weights_81_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_81_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_81_cast = mul(x = var_3670_cast, y = _inversed_attn_weights_81_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_81_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_201_cast = add(x = _inversed_attn_weights_81_cast, y = attention_mask_cast)[name = tensor<string, []>("input_201_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_3674_cast = softmax(axis = var_10, x = input_201_cast)[name = tensor<string, []>("op_3674_cast")];
            tensor<bool, []> attn_output_81_transpose_x_0 = const()[name = tensor<string, []>("attn_output_81_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_81_transpose_y_0 = const()[name = tensor<string, []>("attn_output_81_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_81_cast = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = var_3674_cast, y = value_states_83_cast)[name = tensor<string, []>("attn_output_81_cast")];
            tensor<int32, [4]> var_3677_perm_0 = const()[name = tensor<string, []>("op_3677_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3679 = const()[name = tensor<string, []>("op_3679"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_5 = transpose(perm = var_3677_perm_0, x = attn_output_81_cast)[name = tensor<string, []>("transpose_5")];
            tensor<fp16, [1, 128, 2048]> input_203_cast = reshape(shape = var_3679, x = transpose_5)[name = tensor<string, []>("input_203_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_20_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(476017408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478114624))), name = tensor<string, []>("model_model_layers_20_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_143_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_20_self_attn_o_proj_weight_to_fp16_palettized, x = input_203_cast)[name = tensor<string, []>("linear_143_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_621_cast = add(x = hidden_states_601_cast, y = linear_143_cast)[name = tensor<string, []>("hidden_states_621_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_41 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_41"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3686_cast = pow(x = hidden_states_621_cast, y = var_13_promoted_to_fp16_41)[name = tensor<string, []>("op_3686_cast")];
            tensor<int32, [1]> var_3687 = const()[name = tensor<string, []>("op_3687"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_83_cast = reduce_mean(axes = var_3687, keep_dims = var_23, x = var_3686_cast)[name = tensor<string, []>("variance_83_cast")];
            tensor<fp16, []> var_3689_to_fp16 = const()[name = tensor<string, []>("op_3689_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3690_cast = add(x = variance_83_cast, y = var_3689_to_fp16)[name = tensor<string, []>("op_3690_cast")];
            tensor<fp16, []> var_3691_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3691_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3691_cast = rsqrt(epsilon = var_3691_epsilon_0_to_fp16, x = var_3690_cast)[name = tensor<string, []>("op_3691_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_625_cast = mul(x = hidden_states_621_cast, y = var_3691_cast)[name = tensor<string, []>("hidden_states_625_cast")];
            tensor<fp16, [2048]> model_model_layers_20_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_20_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478114752)))];
            tensor<fp16, [1, 128, 2048]> input_205_cast = mul(x = model_model_layers_20_post_attention_layernorm_weight_to_fp16, y = hidden_states_625_cast)[name = tensor<string, []>("input_205_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_20_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(478118912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(483886144))), name = tensor<string, []>("model_model_layers_20_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_144_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_20_mlp_gate_proj_weight_to_fp16_palettized, x = input_205_cast)[name = tensor<string, []>("linear_144_cast")];
            tensor<fp16, [1, 128, 5632]> var_3703_cast = silu(x = linear_144_cast)[name = tensor<string, []>("op_3703_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_20_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(483886272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489653504))), name = tensor<string, []>("model_model_layers_20_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_145_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_20_mlp_up_proj_weight_to_fp16_palettized, x = input_205_cast)[name = tensor<string, []>("linear_145_cast")];
            tensor<fp16, [1, 128, 5632]> input_209_cast = mul(x = var_3703_cast, y = linear_145_cast)[name = tensor<string, []>("input_209_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_20_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489653632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(495420864))), name = tensor<string, []>("model_model_layers_20_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_146_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_20_mlp_down_proj_weight_to_fp16_palettized, x = input_209_cast)[name = tensor<string, []>("linear_146_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_631_cast = add(x = hidden_states_621_cast, y = linear_146_cast)[name = tensor<string, []>("hidden_states_631_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_42 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_42"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3716_cast = pow(x = hidden_states_631_cast, y = var_13_promoted_to_fp16_42)[name = tensor<string, []>("op_3716_cast")];
            tensor<int32, [1]> var_3717 = const()[name = tensor<string, []>("op_3717"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_85_cast = reduce_mean(axes = var_3717, keep_dims = var_23, x = var_3716_cast)[name = tensor<string, []>("variance_85_cast")];
            tensor<fp16, []> var_3719_to_fp16 = const()[name = tensor<string, []>("op_3719_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3720_cast = add(x = variance_85_cast, y = var_3719_to_fp16)[name = tensor<string, []>("op_3720_cast")];
            tensor<fp16, []> var_3721_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3721_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3721_cast = rsqrt(epsilon = var_3721_epsilon_0_to_fp16, x = var_3720_cast)[name = tensor<string, []>("op_3721_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_635_cast = mul(x = hidden_states_631_cast, y = var_3721_cast)[name = tensor<string, []>("hidden_states_635_cast")];
            tensor<fp16, [2048]> model_model_layers_21_input_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_21_input_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(495420992)))];
            tensor<fp16, [1, 128, 2048]> hidden_states_639_cast = mul(x = model_model_layers_21_input_layernorm_weight_to_fp16, y = hidden_states_635_cast)[name = tensor<string, []>("hidden_states_639_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_21_self_attn_q_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(495425152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497522368))), name = tensor<string, []>("model_model_layers_21_self_attn_q_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_147_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_21_self_attn_q_proj_weight_to_fp16_palettized, x = hidden_states_639_cast)[name = tensor<string, []>("linear_147_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_21_self_attn_k_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497522496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497784704))), name = tensor<string, []>("model_model_layers_21_self_attn_k_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_148_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_21_self_attn_k_proj_weight_to_fp16_palettized, x = hidden_states_639_cast)[name = tensor<string, []>("linear_148_cast")];
            tensor<fp16, [256, 2048]> model_model_layers_21_self_attn_v_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [262144]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(497784832))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498047040))), name = tensor<string, []>("model_model_layers_21_self_attn_v_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([256, 2048])];
            tensor<fp16, [1, 128, 256]> linear_149_cast = linear(bias = linear_1_bias_0_to_fp16, weight = model_model_layers_21_self_attn_v_proj_weight_to_fp16_palettized, x = hidden_states_639_cast)[name = tensor<string, []>("linear_149_cast")];
            tensor<int32, [4]> var_3741 = const()[name = tensor<string, []>("op_3741"), val = tensor<int32, [4]>([1, 128, 32, 64])];
            tensor<fp16, [1, 128, 32, 64]> var_3742_cast = reshape(shape = var_3741, x = linear_147_cast)[name = tensor<string, []>("op_3742_cast")];
            tensor<int32, [4]> q_perm_0 = const()[name = tensor<string, []>("q_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3744 = const()[name = tensor<string, []>("op_3744"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3745_cast = reshape(shape = var_3744, x = linear_148_cast)[name = tensor<string, []>("op_3745_cast")];
            tensor<int32, [4]> key_states_129_perm_0 = const()[name = tensor<string, []>("key_states_129_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> var_3747 = const()[name = tensor<string, []>("op_3747"), val = tensor<int32, [4]>([1, 128, 4, 64])];
            tensor<fp16, [1, 128, 4, 64]> var_3748_cast = reshape(shape = var_3747, x = linear_149_cast)[name = tensor<string, []>("op_3748_cast")];
            tensor<int32, [4]> hidden_states_645_perm_0 = const()[name = tensor<string, []>("hidden_states_645_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 32, 128, 64]> transpose_4 = transpose(perm = q_perm_0, x = var_3742_cast)[name = tensor<string, []>("transpose_4")];
            tensor<fp16, [1, 32, 128, 64]> var_3774_cast = mul(x = transpose_4, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3774_cast")];
            tensor<int32, [4]> x1_85_begin_0 = const()[name = tensor<string, []>("x1_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_85_end_0 = const()[name = tensor<string, []>("x1_85_end_0"), val = tensor<int32, [4]>([1, 32, 128, 32])];
            tensor<bool, [4]> x1_85_end_mask_0 = const()[name = tensor<string, []>("x1_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 32, 128, 32]> x1_85_cast = slice_by_index(begin = x1_85_begin_0, end = x1_85_end_0, end_mask = x1_85_end_mask_0, x = transpose_4)[name = tensor<string, []>("x1_85_cast")];
            tensor<int32, [4]> x2_85_begin_0 = const()[name = tensor<string, []>("x2_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_85_end_0 = const()[name = tensor<string, []>("x2_85_end_0"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<bool, [4]> x2_85_end_mask_0 = const()[name = tensor<string, []>("x2_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 32, 128, 32]> x2_85_cast = slice_by_index(begin = x2_85_begin_0, end = x2_85_end_0, end_mask = x2_85_end_mask_0, x = transpose_4)[name = tensor<string, []>("x2_85_cast")];
            tensor<fp16, []> const_368_promoted_to_fp16 = const()[name = tensor<string, []>("const_368_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 32, 128, 32]> var_3785_cast = mul(x = x2_85_cast, y = const_368_promoted_to_fp16)[name = tensor<string, []>("op_3785_cast")];
            tensor<bool, []> var_3787_interleave_0 = const()[name = tensor<string, []>("op_3787_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> var_3787_cast = concat(axis = var_10, interleave = var_3787_interleave_0, values = (var_3785_cast, x1_85_cast))[name = tensor<string, []>("op_3787_cast")];
            tensor<fp16, [1, 32, 128, 64]> var_3788_cast = mul(x = var_3787_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3788_cast")];
            tensor<fp16, [1, 32, 128, 64]> query_states_cast = add(x = var_3774_cast, y = var_3788_cast)[name = tensor<string, []>("query_states_cast")];
            tensor<fp16, [1, 4, 128, 64]> transpose_3 = transpose(perm = key_states_129_perm_0, x = var_3745_cast)[name = tensor<string, []>("transpose_3")];
            tensor<fp16, [1, 4, 128, 64]> var_3790_cast = mul(x = transpose_3, y = cos_1_to_fp16_palettized)[name = tensor<string, []>("op_3790_cast")];
            tensor<int32, [4]> x1_begin_0 = const()[name = tensor<string, []>("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = tensor<string, []>("x1_end_0"), val = tensor<int32, [4]>([1, 4, 128, 32])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = tensor<string, []>("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 4, 128, 32]> x1_cast = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = transpose_3)[name = tensor<string, []>("x1_cast")];
            tensor<int32, [4]> x2_begin_0 = const()[name = tensor<string, []>("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_end_0 = const()[name = tensor<string, []>("x2_end_0"), val = tensor<int32, [4]>([1, 4, 128, 64])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = tensor<string, []>("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 128, 32]> x2_cast = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = transpose_3)[name = tensor<string, []>("x2_cast")];
            tensor<fp16, []> const_371_promoted_to_fp16 = const()[name = tensor<string, []>("const_371_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [1, 4, 128, 32]> var_3801_cast = mul(x = x2_cast, y = const_371_promoted_to_fp16)[name = tensor<string, []>("op_3801_cast")];
            tensor<bool, []> var_3803_interleave_0 = const()[name = tensor<string, []>("op_3803_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 4, 128, 64]> var_3803_cast = concat(axis = var_10, interleave = var_3803_interleave_0, values = (var_3801_cast, x1_cast))[name = tensor<string, []>("op_3803_cast")];
            tensor<fp16, [1, 4, 128, 64]> var_3804_cast = mul(x = var_3803_cast, y = sin_1_to_fp16_palettized)[name = tensor<string, []>("op_3804_cast")];
            tensor<fp16, [1, 4, 128, 64]> hidden_states_641_cast = add(x = var_3790_cast, y = var_3804_cast)[name = tensor<string, []>("hidden_states_641_cast")];
            tensor<int32, [1]> var_3813_axes_0 = const()[name = tensor<string, []>("op_3813_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3813_cast = expand_dims(axes = var_3813_axes_0, x = hidden_states_641_cast)[name = tensor<string, []>("op_3813_cast")];
            tensor<int32, [5]> hidden_states_643_reps_0 = const()[name = tensor<string, []>("hidden_states_643_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_643_cast = tile(reps = hidden_states_643_reps_0, x = var_3813_cast)[name = tensor<string, []>("hidden_states_643_cast")];
            tensor<int32, [4]> var_3821 = const()[name = tensor<string, []>("op_3821"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> key_states_cast = reshape(shape = var_3821, x = hidden_states_643_cast)[name = tensor<string, []>("key_states_cast")];
            tensor<int32, [1]> var_3830_axes_0 = const()[name = tensor<string, []>("op_3830_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 128, 64]> transpose_2 = transpose(perm = hidden_states_645_perm_0, x = var_3748_cast)[name = tensor<string, []>("transpose_2")];
            tensor<fp16, [1, 4, 1, 128, 64]> var_3830_cast = expand_dims(axes = var_3830_axes_0, x = transpose_2)[name = tensor<string, []>("op_3830_cast")];
            tensor<int32, [5]> hidden_states_647_reps_0 = const()[name = tensor<string, []>("hidden_states_647_reps_0"), val = tensor<int32, [5]>([1, 1, 8, 1, 1])];
            tensor<fp16, [1, 4, 8, 128, 64]> hidden_states_647_cast = tile(reps = hidden_states_647_reps_0, x = var_3830_cast)[name = tensor<string, []>("hidden_states_647_cast")];
            tensor<int32, [4]> var_3838 = const()[name = tensor<string, []>("op_3838"), val = tensor<int32, [4]>([1, 32, 128, 64])];
            tensor<fp16, [1, 32, 128, 64]> value_states_cast = reshape(shape = var_3838, x = hidden_states_647_cast)[name = tensor<string, []>("value_states_cast")];
            tensor<int32, [4]> var_3840_perm_0 = const()[name = tensor<string, []>("op_3840_perm_0"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<bool, []> var_3841_transpose_x_0 = const()[name = tensor<string, []>("op_3841_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_3841_transpose_y_0 = const()[name = tensor<string, []>("op_3841_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 64, 128]> transpose_1 = transpose(perm = var_3840_perm_0, x = key_states_cast)[name = tensor<string, []>("transpose_1")];
            tensor<fp16, [1, 32, 128, 128]> var_3841_cast = matmul(transpose_x = var_3841_transpose_x_0, transpose_y = var_3841_transpose_y_0, x = query_states_cast, y = transpose_1)[name = tensor<string, []>("op_3841_cast")];
            tensor<fp16, []> _inversed_attn_weights_85_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_attn_weights_85_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 32, 128, 128]> _inversed_attn_weights_85_cast = mul(x = var_3841_cast, y = _inversed_attn_weights_85_y_0_to_fp16)[name = tensor<string, []>("_inversed_attn_weights_85_cast")];
            tensor<fp16, [1, 32, 128, 128]> input_211_cast = add(x = _inversed_attn_weights_85_cast, y = attention_mask_cast)[name = tensor<string, []>("input_211_cast")];
            tensor<fp16, [1, 32, 128, 128]> var_3845_cast = softmax(axis = var_10, x = input_211_cast)[name = tensor<string, []>("op_3845_cast")];
            tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 32, 128, 64]> attn_output_85_cast = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = var_3845_cast, y = value_states_cast)[name = tensor<string, []>("attn_output_85_cast")];
            tensor<int32, [4]> var_3848_perm_0 = const()[name = tensor<string, []>("op_3848_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3850 = const()[name = tensor<string, []>("op_3850"), val = tensor<int32, [3]>([1, 128, 2048])];
            tensor<fp16, [1, 128, 32, 64]> transpose_0 = transpose(perm = var_3848_perm_0, x = attn_output_85_cast)[name = tensor<string, []>("transpose_0")];
            tensor<fp16, [1, 128, 2048]> input_213_cast = reshape(shape = var_3850, x = transpose_0)[name = tensor<string, []>("input_213_cast")];
            tensor<fp16, [2048, 2048]> model_model_layers_21_self_attn_o_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [2097152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498047168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500144384))), name = tensor<string, []>("model_model_layers_21_self_attn_o_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 2048])];
            tensor<fp16, [1, 128, 2048]> linear_150_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_21_self_attn_o_proj_weight_to_fp16_palettized, x = input_213_cast)[name = tensor<string, []>("linear_150_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_651_cast = add(x = hidden_states_631_cast, y = linear_150_cast)[name = tensor<string, []>("hidden_states_651_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_43 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_43"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3857_cast = pow(x = hidden_states_651_cast, y = var_13_promoted_to_fp16_43)[name = tensor<string, []>("op_3857_cast")];
            tensor<int32, [1]> var_3858 = const()[name = tensor<string, []>("op_3858"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_87_cast = reduce_mean(axes = var_3858, keep_dims = var_23, x = var_3857_cast)[name = tensor<string, []>("variance_87_cast")];
            tensor<fp16, []> var_3860_to_fp16 = const()[name = tensor<string, []>("op_3860_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3861_cast = add(x = variance_87_cast, y = var_3860_to_fp16)[name = tensor<string, []>("op_3861_cast")];
            tensor<fp16, []> var_3862_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3862_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3862_cast = rsqrt(epsilon = var_3862_epsilon_0_to_fp16, x = var_3861_cast)[name = tensor<string, []>("op_3862_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_655_cast = mul(x = hidden_states_651_cast, y = var_3862_cast)[name = tensor<string, []>("hidden_states_655_cast")];
            tensor<fp16, [2048]> model_model_layers_21_post_attention_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_layers_21_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500144512)))];
            tensor<fp16, [1, 128, 2048]> input_215_cast = mul(x = model_model_layers_21_post_attention_layernorm_weight_to_fp16, y = hidden_states_655_cast)[name = tensor<string, []>("input_215_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_21_mlp_gate_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(500148672))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505915904))), name = tensor<string, []>("model_model_layers_21_mlp_gate_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_151_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_21_mlp_gate_proj_weight_to_fp16_palettized, x = input_215_cast)[name = tensor<string, []>("linear_151_cast")];
            tensor<fp16, [1, 128, 5632]> var_3874_cast = silu(x = linear_151_cast)[name = tensor<string, []>("op_3874_cast")];
            tensor<fp16, [5632, 2048]> model_model_layers_21_mlp_up_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505916032))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511683264))), name = tensor<string, []>("model_model_layers_21_mlp_up_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([5632, 2048])];
            tensor<fp16, [1, 128, 5632]> linear_152_cast = linear(bias = linear_4_bias_0_to_fp16_palettized, weight = model_model_layers_21_mlp_up_proj_weight_to_fp16_palettized, x = input_215_cast)[name = tensor<string, []>("linear_152_cast")];
            tensor<fp16, [1, 128, 5632]> input_219_cast = mul(x = var_3874_cast, y = linear_152_cast)[name = tensor<string, []>("input_219_cast")];
            tensor<fp16, [2048, 5632]> model_model_layers_21_mlp_down_proj_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [5767168]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(511683392))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517450624))), name = tensor<string, []>("model_model_layers_21_mlp_down_proj_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([2048, 5632])];
            tensor<fp16, [1, 128, 2048]> linear_153_cast = linear(bias = linear_0_bias_0_to_fp16, weight = model_model_layers_21_mlp_down_proj_weight_to_fp16_palettized, x = input_219_cast)[name = tensor<string, []>("linear_153_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_661_cast = add(x = hidden_states_651_cast, y = linear_153_cast)[name = tensor<string, []>("hidden_states_661_cast")];
            tensor<fp16, []> var_13_promoted_to_fp16_44 = const()[name = tensor<string, []>("op_13_promoted_to_fp16_44"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1, 128, 2048]> var_3883_cast = pow(x = hidden_states_661_cast, y = var_13_promoted_to_fp16_44)[name = tensor<string, []>("op_3883_cast")];
            tensor<int32, [1]> var_3884 = const()[name = tensor<string, []>("op_3884"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 128, 1]> variance_cast = reduce_mean(axes = var_3884, keep_dims = var_23, x = var_3883_cast)[name = tensor<string, []>("variance_cast")];
            tensor<fp16, []> var_3886_to_fp16 = const()[name = tensor<string, []>("op_3886_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 128, 1]> var_3887_cast = add(x = variance_cast, y = var_3886_to_fp16)[name = tensor<string, []>("op_3887_cast")];
            tensor<fp16, []> var_3888_epsilon_0_to_fp16 = const()[name = tensor<string, []>("op_3888_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)];
            tensor<fp16, [1, 128, 1]> var_3888_cast = rsqrt(epsilon = var_3888_epsilon_0_to_fp16, x = var_3887_cast)[name = tensor<string, []>("op_3888_cast")];
            tensor<fp16, [1, 128, 2048]> hidden_states_665_cast = mul(x = hidden_states_661_cast, y = var_3888_cast)[name = tensor<string, []>("hidden_states_665_cast")];
            tensor<fp16, [2048]> model_model_norm_weight_to_fp16 = const()[name = tensor<string, []>("model_model_norm_weight_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517450752)))];
            tensor<fp16, [1, 128, 2048]> input_cast = mul(x = model_model_norm_weight_to_fp16, y = hidden_states_665_cast)[name = tensor<string, []>("input_cast")];
            tensor<fp16, [32003, 2048]> model_lm_head_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [32771072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517454912))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550226048))), name = tensor<string, []>("model_lm_head_weight_to_fp16_palettized"), shape = tensor<uint32, [2]>([32003, 2048])];
            tensor<fp16, [32003]> linear_154_bias_0_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [16002]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550226176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(550242304))), name = tensor<string, []>("linear_154_bias_0_to_fp16_palettized"), shape = tensor<uint32, [1]>([32003])];
            tensor<fp16, [1, 128, 32003]> linear_154_cast = linear(bias = linear_154_bias_0_to_fp16_palettized, weight = model_lm_head_weight_to_fp16_palettized, x = input_cast)[name = tensor<string, []>("linear_154_cast")];
            tensor<string, []> linear_154_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("linear_154_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 128, 32003]> logits = cast(dtype = linear_154_cast_to_fp32_dtype_0, x = linear_154_cast)[name = tensor<string, []>("cast_0")];
        } -> (logits);
}