init

config.json

@@ -0,0 +1,58 @@
+{
+  "architectures": [
+    "HFOpenMoeForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoModel": "modeling_hf_openmoe.HFOpenMoeModel",
+    "AutoConfig": "configuration_hf_openmoe.HFOpenMoeConfig",
+    "AutoModelForCausalLM": "modeling_hf_openmoe.HFOpenMoeForCausalLM"
+  },
+  "bos_token_id": 0,
+  "dropout_rate": 0.0,
+  "enable_comm_overlap": false,
+  "enable_hierarchical_alltoall": false,
+  "enable_kernel": false,
+  "enable_load_balance": false,
+  "eos_token_id": 1,
+  "expert_parallel": null,
+  "head_dim": 128,
+  "hidden_act": "swiglu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "label_smoothing": 0.001,
+  "layer_norm_epsilon": 1e-06,
+  "load_balance_beam_width": 8,
+  "load_balance_group_swap_factor": 0.4,
+  "load_balance_tolerance": 0.1,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "mlp_gated": true,
+  "model_type": "openmoe",
+  "moe_layer_interval": 6,
+  "num_attention_heads": 24,
+  "num_experts": 32,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 24,
+  "pad_token_id": 0,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "router_aux_loss_factor": 0.01,
+  "router_capacity_factor_eval": 2.0,
+  "router_capacity_factor_train": 1.25,
+  "router_drop_tks": true,
+  "router_min_capacity": 4,
+  "router_noisy_policy": null,
+  "router_topk": 2,
+  "router_z_loss_factor": 0.0001,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.43.3",
+  "use_cache": true,
+  "vocab_size": 256384,
+  "z_loss_factor": 0.01
+}

configuration_hf_openmoe.py

@@ -0,0 +1,53 @@
+from transformers import LlamaConfig
+
+__all__ = ["HFOpenMoeConfig"]
+
+
+class HFOpenMoeConfig(LlamaConfig):
+    model_type = "openmoe"
+    def __init__(
+            self,
+            num_experts: int = 32,
+            moe_layer_interval: int = 6,
+            router_topk: int = 2,
+            router_capacity_factor_train: float = 1.25,
+            router_capacity_factor_eval: float = 2.0,
+            router_min_capacity: int = 4,
+            router_noisy_policy: str = None,
+            router_drop_tks: bool = True,
+            router_aux_loss_factor: float = 0.01,
+            router_z_loss_factor: float = 0.0001,
+            mlp_gated: bool = True,
+            label_smoothing: float = 0.001,
+            z_loss_factor: float = 0.01,
+            enable_load_balance: bool = False,
+            load_balance_tolerance: float = 0.1,
+            load_balance_beam_width: int = 8,
+            load_balance_group_swap_factor: float = 0.4,
+            enable_kernel: bool = False,
+            enable_comm_overlap: bool = False,
+            enable_hierarchical_alltoall: bool = False,
+            **kwargs
+    ):
+        self.num_experts = num_experts
+        self.moe_layer_interval = moe_layer_interval
+        self.router_topk = router_topk
+        self.router_capacity_factor_train = router_capacity_factor_train
+        self.router_capacity_factor_eval = router_capacity_factor_eval
+        self.router_min_capacity = router_min_capacity
+        self.router_noisy_policy = router_noisy_policy
+        self.router_drop_tks = router_drop_tks
+        self.router_aux_loss_factor = router_aux_loss_factor
+        self.router_z_loss_factor = router_z_loss_factor
+        self.mlp_gated = mlp_gated
+        self.label_smoothing = label_smoothing
+        self.z_loss_factor = z_loss_factor
+        self.enable_load_balance = enable_load_balance
+        self.load_balance_tolerance = load_balance_tolerance
+        self.load_balance_beam_width = load_balance_beam_width
+        self.load_balance_group_swap_factor = load_balance_group_swap_factor
+        self.enable_kernel = enable_kernel
+        self.enable_comm_overlap = enable_comm_overlap
+        self.enable_hierarchical_alltoall = enable_hierarchical_alltoall
+
+        super().__init__(**kwargs)

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "pad_token_id": 0,
+  "transformers_version": "4.43.3"
+}

model-00001-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec229bf56b0726cbaba3e67994d6c96629d5d7f79bf4ba94bab28e87b41ac3dc
+size 4986346648

model-00002-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb5f12a244b2e8c751852f7a844cd980f7733dca6d0de96c35b115a6d1bc07c9
+size 4966062856

model-00003-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67cddbb2160baa44e38bbfae81eecc27c080a424fcffa0e4f06bcd26dc2c2ac4
+size 4899325848

model-00004-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf85eb0992701ecdb8feb5fffea136955ba2c896bfd07241995bb2fb2977bd7a
+size 4966062776

model-00005-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c58b7a5cb906ad76623f2b5eff4ef912d928a69109954bf5d836c7632247279
+size 4999676432

model-00006-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:106916f17274c0b557de3f4e3976af5085fb21c9e5d2413ecc3d281d0b06f40f
+size 4999930240

model-00007-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76a36e031c80311276ad0c4135be8f49a8603790e5b9ac64f8ffb944051335d8
+size 4999642592

model-00008-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:156af9a8472948caa702719c7454f61eeea897ead29c1a6a74540749bc2afb56
+size 4999964088

model-00009-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:da28ce83a95c60d0cfd4da82921f2b3249e49c02af147aad8874e9365a0d0ad9
+size 4966062928

model-00010-of-00010.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab30a6648ad25187f2a104526c316f0e49d010b5647812eb9c929c357cb73b01
+size 2637186096

model.safetensors.index.json

@@ -0,0 +1,618 @@
+{
+  "metadata": {
+    "total_size": 47420186624
+  },
+  "weight_map": {
+    "lm_head.weight": "model-00010-of-00010.safetensors",
+    "model.embed_tokens.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.mlp.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.mlp.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.mlp.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.0.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.mlp.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.mlp.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.mlp.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.1.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.10.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.10.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.extra_mlp.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.extra_mlp.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.extra_mlp.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.mlp.experts.0.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.0.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.0.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.1.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.1.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.1.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.10.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.10.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.10.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.11.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.11.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.11.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.12.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.12.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.12.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.13.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.13.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.13.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.14.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.14.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.14.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.15.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.15.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.15.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.16.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.16.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.16.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.17.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.17.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.17.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.18.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.18.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.18.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.19.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.19.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.19.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.2.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.2.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.2.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.20.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.20.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.20.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.21.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.21.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.21.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.22.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.22.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.22.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.23.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.23.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.23.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.24.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.24.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.24.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.25.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.25.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.25.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.26.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.26.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.26.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.27.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.27.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.27.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.28.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.28.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.28.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.29.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.29.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.29.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.3.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.3.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.3.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.30.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.30.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.30.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.31.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.31.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.31.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.mlp.experts.4.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.4.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.4.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.5.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.5.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.5.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.6.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.6.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.6.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.7.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.7.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.7.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.8.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.8.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.8.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.9.down_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.9.gate_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.experts.9.up_proj.weight": "model-00004-of-00010.safetensors",
+    "model.layers.11.mlp.gate.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.pre_extra_mlp_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.11.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.11.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.12.input_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.mlp.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.mlp.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.mlp.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.post_attention_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.self_attn.k_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.self_attn.o_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.self_attn.q_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.12.self_attn.v_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.input_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.mlp.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.mlp.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.mlp.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.post_attention_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.self_attn.k_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.self_attn.o_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.self_attn.q_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.13.self_attn.v_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.input_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.mlp.down_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.mlp.gate_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.mlp.up_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.post_attention_layernorm.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.self_attn.k_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.self_attn.o_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.self_attn.q_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.14.self_attn.v_proj.weight": "model-00005-of-00010.safetensors",
+    "model.layers.15.input_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.mlp.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.mlp.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.mlp.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.post_attention_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.self_attn.k_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.self_attn.o_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.self_attn.q_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.15.self_attn.v_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.input_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.mlp.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.mlp.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.mlp.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.post_attention_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.self_attn.k_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.self_attn.o_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.self_attn.q_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.16.self_attn.v_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.extra_mlp.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.extra_mlp.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.extra_mlp.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.input_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.0.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.0.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.0.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.1.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.1.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.1.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.10.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.10.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.10.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.11.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.11.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.11.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.12.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.12.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.12.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.13.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.13.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.13.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.14.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.14.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.14.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.15.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.15.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.15.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.16.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.16.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.16.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.17.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.17.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.17.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.18.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.18.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.18.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.19.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.19.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.19.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.2.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.2.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.2.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.20.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.20.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.20.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.21.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.21.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.21.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.22.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.22.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.22.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.23.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.23.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.23.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.24.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.24.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.24.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.25.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.25.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.25.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.26.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.26.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.26.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.27.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.27.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.27.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.28.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.28.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.28.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.29.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.29.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.29.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.3.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.3.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.3.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.30.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.30.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.30.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.31.down_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.31.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.31.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.mlp.experts.4.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.4.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.4.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.5.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.5.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.5.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.6.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.6.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.6.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.7.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.7.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.7.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.8.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.8.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.8.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.9.down_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.9.gate_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.experts.9.up_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.mlp.gate.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.post_attention_layernorm.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.pre_extra_mlp_layernorm.weight": "model-00007-of-00010.safetensors",
+    "model.layers.17.self_attn.k_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.self_attn.o_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.self_attn.q_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.17.self_attn.v_proj.weight": "model-00006-of-00010.safetensors",
+    "model.layers.18.input_layernorm.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.mlp.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.18.mlp.gate_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.mlp.up_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.post_attention_layernorm.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.self_attn.k_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.self_attn.o_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.self_attn.q_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.18.self_attn.v_proj.weight": "model-00007-of-00010.safetensors",
+    "model.layers.19.input_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.mlp.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.mlp.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.mlp.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.post_attention_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.self_attn.k_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.self_attn.o_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.self_attn.q_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.19.self_attn.v_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.2.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.mlp.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.mlp.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.mlp.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.2.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.20.input_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.mlp.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.mlp.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.mlp.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.post_attention_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.self_attn.k_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.self_attn.o_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.self_attn.q_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.20.self_attn.v_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.input_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.mlp.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.mlp.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.mlp.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.post_attention_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.self_attn.k_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.self_attn.o_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.self_attn.q_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.21.self_attn.v_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.input_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.mlp.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.mlp.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.mlp.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.post_attention_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.self_attn.k_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.self_attn.o_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.self_attn.q_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.22.self_attn.v_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.extra_mlp.down_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.extra_mlp.gate_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.extra_mlp.up_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.input_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.0.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.0.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.0.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.1.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.1.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.1.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.10.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.10.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.10.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.11.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.11.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.11.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.12.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.12.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.12.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.13.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.13.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.13.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.14.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.14.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.14.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.15.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.15.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.15.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.16.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.16.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.16.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.17.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.17.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.17.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.18.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.18.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.18.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.19.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.19.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.19.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.2.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.2.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.2.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.20.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.20.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.20.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.21.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.21.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.21.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.22.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.22.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.22.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.23.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.23.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.23.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.24.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.24.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.24.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.25.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.25.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.25.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.26.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.26.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.26.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.27.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.27.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.27.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.28.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.28.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.28.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.29.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.29.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.29.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.3.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.3.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.3.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.30.down_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.30.gate_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.30.up_proj.weight": "model-00009-of-00010.safetensors",
+    "model.layers.23.mlp.experts.31.down_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.mlp.experts.31.gate_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.mlp.experts.31.up_proj.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.mlp.experts.4.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.4.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.4.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.5.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.5.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.5.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.6.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.6.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.6.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.7.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.7.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.7.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.8.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.8.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.8.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.9.down_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.9.gate_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.experts.9.up_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.mlp.gate.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.post_attention_layernorm.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.pre_extra_mlp_layernorm.weight": "model-00010-of-00010.safetensors",
+    "model.layers.23.self_attn.k_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.self_attn.o_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.self_attn.q_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.23.self_attn.v_proj.weight": "model-00008-of-00010.safetensors",
+    "model.layers.3.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.mlp.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.mlp.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.mlp.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.3.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.mlp.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.mlp.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.mlp.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.4.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.extra_mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.extra_mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.extra_mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.input_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.0.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.0.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.0.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.1.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.1.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.1.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.10.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.10.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.10.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.11.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.11.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.11.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.12.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.12.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.12.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.13.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.13.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.13.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.14.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.14.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.14.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.15.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.15.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.15.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.16.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.16.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.16.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.17.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.17.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.17.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.18.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.18.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.18.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.19.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.19.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.19.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.2.down_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.2.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.2.up_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.20.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.20.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.20.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.21.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.21.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.21.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.22.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.22.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.22.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.23.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.23.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.23.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.24.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.24.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.24.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.25.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.25.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.25.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.26.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.26.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.26.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.27.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.27.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.27.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.28.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.28.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.28.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.29.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.29.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.29.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.3.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.3.gate_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.mlp.experts.3.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.30.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.30.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.30.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.31.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.31.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.31.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.mlp.experts.4.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.4.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.4.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.5.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.5.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.5.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.6.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.6.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.6.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.7.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.7.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.7.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.8.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.8.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.8.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.9.down_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.9.gate_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.experts.9.up_proj.weight": "model-00002-of-00010.safetensors",
+    "model.layers.5.mlp.gate.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.post_attention_layernorm.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.pre_extra_mlp_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.5.self_attn.k_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.self_attn.o_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.self_attn.q_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.5.self_attn.v_proj.weight": "model-00001-of-00010.safetensors",
+    "model.layers.6.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.6.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.7.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.8.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.input_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.mlp.down_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.mlp.gate_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.mlp.up_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.post_attention_layernorm.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.self_attn.k_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.self_attn.o_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.self_attn.q_proj.weight": "model-00003-of-00010.safetensors",
+    "model.layers.9.self_attn.v_proj.weight": "model-00003-of-00010.safetensors",
+    "model.norm.weight": "model-00010-of-00010.safetensors"
+  }
+}

modeling_hf_openmoe.py

@@ -0,0 +1,1121 @@
+# coding=utf-8
+# Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on EleutherAI's GPT-NeoX library and the GPT-NeoX
+# and OPT implementations in this library. It has been modified from its
+# original forms to accommodate minor architectural differences compared
+# to GPT-NeoX and OPT used by the Meta AI team that trained the model.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch OpenMoE model."""
+import math
+from typing import List, Union
+from typing import Optional, Tuple
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import functional as F
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import (
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    logging,
+    replace_return_docstrings,
+)
+
+# from .llama_attn import LlamaAttention
+from .configuration_hf_openmoe import HFOpenMoeConfig
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "HFOpenMoeConfig"
+
+
+# Copied from transformers.models.bart.modeling_bart._make_causal_mask
+def _make_causal_mask(
+        input_ids_shape: torch.Size, dtype: torch.dtype, device: torch.device, past_key_values_length: int = 0
+):
+    """
+    Make causal mask used for bi-directional self-attention.
+    """
+    bsz, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.finfo(dtype).min, device=device)
+    mask_cond = torch.arange(mask.size(-1), device=device)
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+
+    if past_key_values_length > 0:
+        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
+    return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+# Copied from transformers.models.bart.modeling_bart._expand_mask
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """
+    Expands attention_mask from `[bsz, seq_len]` to `[bsz, 1, tgt_seq_len, src_seq_len]`.
+    """
+    bsz, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(bsz, 1, tgt_len, src_len).to(dtype)
+
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(inverted_mask.to(torch.bool), torch.finfo(dtype).min)
+
+
+def apply_rotary_embedding(q, k, cos, sin, decode=False, rotary_index=None):
+    # q:  (bs, q_len, num_heads, head_dim)
+    # k:  (bs, q_len [+past_kv_len], num_heads, head_dim)
+    # cos: (max_seq_len, head_dim)
+    # sin: (max_seq_len, head_dim)
+    # rotary_index: (bs, 1)  # only used during decoding, when one query token is input at a time
+    """Helper function to apply Rotary Embeddings."""
+    cos = cos.to(q.dtype)
+    sin = sin.to(q.dtype)
+
+    if len(k.shape) == 3:  # for multi query attention
+        k = k.unsqueeze(2)
+        multiquery = True
+    else:
+        multiquery = False
+
+    batch, qlen, qheads, d = q.shape
+    kbatch, klen, kheads, kd = k.shape
+    assert batch == kbatch, f"{batch} != {kbatch}"
+    assert d == kd, f"{d} != {kd}"
+    if decode and qlen == 1 and rotary_index is not None:
+        qcos = cos[rotary_index, :]  # (bs, 1, head_dim)
+        qsin = sin[rotary_index, :]  # (bs, 1, head_dim)
+        qcos = qcos.unsqueeze(2)  # (bs, q_len=1, 1, head_dim)  # broadcast to all heads
+        qsin = qsin.unsqueeze(2)  # (bs, q_len=1, 1, head_dim)
+    else:
+        qcos, qsin = cos[:qlen, :], sin[:qlen, :]  # (q_len, head_dim)
+        qcos = qcos.unsqueeze(0).unsqueeze(2)  # (1, q_len, 1, head_dim)
+        qsin = qsin.unsqueeze(0).unsqueeze(2)
+
+    kcos, ksin = cos[:klen, :], sin[:klen, :]  # (k_len, head_dim)
+    kcos = kcos.unsqueeze(0).unsqueeze(
+        2)  # (1, k_len, 1, head_dim)  # broadcast to the whole batch, broadcast to all heads
+    ksin = ksin.unsqueeze(0).unsqueeze(2)  # (1, k_len, 1, head_dim)
+    out_q = (q * qcos) + (rotate_half(q) * qsin)
+    out_k = (k * kcos) + (rotate_half(k) * ksin)
+
+    if multiquery:
+        out_k = out_k.squeeze(2)
+
+    return out_q, out_k
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2:]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+class LlamaRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        LlamaRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+
+def swiglu_act_fn(x):
+    """Gated linear unit activation function.
+    Args:
+        x : input array
+        axis: the axis along which the split should be computed (default: -1)
+    """
+    size = x.shape[-1]
+    assert size % 2 == 0, "axis size must be divisible by 2"
+    x1, x2 = torch.split(x, size // 2, -1)
+    return x1 * (x2 * torch.sigmoid(x2))
+
+
+class HFOpenMoeMLP(torch.nn.Module):
+    def __init__(self, config: HFOpenMoeConfig):
+        super().__init__()
+        assert config.hidden_act == "swiglu"
+        self.ffn_dim = config.intermediate_size
+        self.hidden_dim = config.hidden_size
+
+        self.gate_proj = nn.Linear(self.hidden_dim, self.ffn_dim * 2, bias=False)
+        self.up_proj = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+        self.down_proj = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)
+
+    def forward(self, hidden_states):
+        return self.down_proj(swiglu_act_fn(self.gate_proj(hidden_states)) * self.up_proj(hidden_states))
+
+
+def moe_cumsum(inputs: torch.Tensor):
+    return torch.cumsum(inputs, dim=0) - 1
+
+
+class HFOpenMoeTop2Router(torch.nn.Module):
+    def __init__(self, config: HFOpenMoeConfig):
+        super().__init__()
+        assert config.router_topk == 2
+        self.k_value = 2
+        self.capacity_factor_train = config.router_capacity_factor_train
+        self.capacity_factor_eval = config.router_capacity_factor_eval
+        self.min_capacity = config.router_min_capacity
+        self.drop_tks = config.router_drop_tks
+
+    def get_capacity(self, logits_shape):
+        capacity_factor = self.capacity_factor_train if self.training else self.capacity_factor_eval
+        capacity = math.floor(self.k_value * capacity_factor * logits_shape[-2] / logits_shape[-1])
+        capacity += capacity % 2
+        capacity = max(capacity, self.min_capacity)
+        assert capacity > 0
+        return int(capacity)
+
+    def forward(self, inputs: torch.Tensor) -> Tuple:
+        assert inputs.dtype == torch.float, "Router input should be FP32"
+
+        probs = F.softmax(inputs, dim=-1)
+        num_experts = probs.size(-1)
+        capacity = self.get_capacity(inputs.shape)
+
+        top1_idx = torch.argmax(probs, dim=-1)
+        mask1 = F.one_hot(top1_idx, num_classes=num_experts).to(torch.int32)
+        logits_except1 = probs.masked_fill(mask1.bool(), float("-inf"))
+        top2_idx = torch.argmax(logits_except1, dim=-1)
+        mask2 = F.one_hot(top2_idx, num_classes=num_experts).to(torch.int32)
+
+        rank1 = moe_cumsum(mask1)  # rank1: [s, e]
+        rank2 = moe_cumsum(mask2)
+        rank2 += torch.sum(mask1, dim=-2, keepdim=True)
+
+        mask1 *= torch.lt(rank1, capacity)
+        mask2 *= torch.lt(rank2, capacity)
+        used_capacity = mask1.sum(dim=0) + mask2.sum(dim=0)
+
+        rank1 = torch.sum(mask1 * rank1, dim=-1)
+        rank2 = torch.sum(mask2 * rank2, dim=-1)
+
+        weight1 = mask1 * probs.type_as(inputs)
+        weight2 = mask2 * probs.type_as(inputs)
+
+        cb_weight = torch.zeros(inputs.shape + (capacity,), device=inputs.device)
+        sec_mask = torch.zeros_like(cb_weight, dtype=torch.bool)
+        indices = torch.arange(0, inputs.shape[0], device=inputs.device)
+        cb_weight[indices, top1_idx[indices], rank1[indices]] += weight1[indices, top1_idx[indices]]
+        cb_weight[indices, top2_idx[indices], rank2[indices]] += weight2[indices, top2_idx[indices]]
+        sec_mask[indices, top1_idx[indices], rank1[indices]] |= mask1.bool()[indices, top1_idx[indices]]
+        sec_mask[indices, top2_idx[indices], rank2[indices]] |= mask2.bool()[indices, top2_idx[indices]]
+
+        return used_capacity, cb_weight, sec_mask
+
+
+class HFOpenMoeSparseMLP(torch.nn.Module):
+    def __init__(self, config: HFOpenMoeConfig):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.num_experts = config.num_experts
+
+        self.gate = torch.nn.Linear(self.hidden_size, config.num_experts, bias=False)
+
+        self.experts = nn.ModuleList([HFOpenMoeMLP(config) for _ in range(self.num_experts)])
+        self.router = HFOpenMoeTop2Router(config)
+
+    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
+        # reshape the input tokens
+        tokens = hidden_states.reshape(-1, self.hidden_size)
+        inputs = hidden_states
+
+        # the data type of the inputs in the gating should be fp32
+        fp32_input = tokens.to(torch.float)
+        self.gate = self.gate.to(torch.float)
+        gate_output = self.gate(fp32_input)
+
+        used_capacity, *route_result_list = self.router(inputs=gate_output)
+
+        sec_mask_f = route_result_list[1].type_as(inputs)
+        dispatch_data = torch.matmul(sec_mask_f.permute(1, 2, 0), tokens)
+
+        expert_output = self._local_process(dispatch_data)
+
+        combine_weights = route_result_list[0].type_as(inputs)
+        combine_weights = combine_weights.view(combine_weights.shape[0], -1)
+        expert_output = expert_output.view(-1, expert_output.shape[-1])
+        ans = torch.matmul(combine_weights, expert_output)
+
+        ans = ans.reshape(inputs.shape)
+        return ans
+
+    def _local_process(self, expert_in: torch.Tensor) -> torch.Tensor:
+        expert_in = expert_in.unsqueeze(0)
+        x = expert_in
+
+        # Copied from colossalai MLPExperts class
+        e = x.size(1)
+        h = x.size(-1)
+
+        x = x.transpose(0, 1)
+        inshape = x.shape
+        x = x.reshape(e, -1, h)
+
+        x = [self.experts[i](x[i]) for i in range(e)]
+
+        x = torch.cat([x[i].unsqueeze(0) for i in range(e)], dim=0)
+        x = x.reshape(inshape)
+        x = x.transpose(0, 1).contiguous()
+
+        expert_out = x
+        return expert_out
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+class HFOpenMoeAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: HFOpenMoeConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = config.head_dim
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.pretraining_tp = config.pretraining_tp
+        self.max_position_embeddings = config.max_position_embeddings
+
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
+        self.generate_fixed_pos_embedding(self.head_dim, self.max_position_embeddings, 1.0, 1e4)
+        self.use_kernel = config.enable_kernel
+
+    def _shape(self, tensor: torch.Tensor, seq_len: int, bsz: int):
+        return tensor.view(bsz, seq_len, self.num_heads, self.head_dim).transpose(1, 2).contiguous()
+
+    def generate_fixed_pos_embedding(self, features, length, min_timescale=1.0, max_timescale=10000.0):
+        """Generate Sin/Cos for Rotary Embeddings.
+
+        Args:
+          features: an integer
+          length: an integer
+          min_timescale: an optional float
+          max_timescale: an optional float
+
+        Returns:
+          output_sin: a float32 Tensor with shape [length, features]
+          output_cos: a float32 Tensor with shape [length, features]
+        """
+        fraction = torch.arange(0, features, 2, dtype=torch.float32) / features
+        timescale = min_timescale * (max_timescale / min_timescale) ** fraction
+        rotational_frequency = 1.0 / timescale
+
+        sinusoid_inp = torch.einsum("i,j->ij", torch.arange(length, dtype=torch.float32), rotational_frequency)
+
+        sinusoid_inp = torch.cat([sinusoid_inp, sinusoid_inp], dim=-1)
+
+        self.register_buffer('sin', torch.sin(sinusoid_inp),
+                             persistent=False)  # persistent=False --> buffer won't appear in the state_dict
+        self.register_buffer('cos', torch.cos(sinusoid_inp), persistent=False)
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: bool = False,
+            use_cache: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        if self.pretraining_tp > 1:
+            key_value_slicing = (self.num_key_value_heads * self.head_dim) // self.pretraining_tp
+            query_slices = self.q_proj.weight.split((self.num_heads * self.head_dim) // self.pretraining_tp, dim=0)
+            key_slices = self.k_proj.weight.split(key_value_slicing, dim=0)
+            value_slices = self.v_proj.weight.split(key_value_slicing, dim=0)
+
+            query_states = [F.linear(hidden_states, query_slices[i]) for i in range(self.pretraining_tp)]
+            query_states = torch.cat(query_states, dim=-1)
+
+            key_states = [F.linear(hidden_states, key_slices[i]) for i in range(self.pretraining_tp)]
+            key_states = torch.cat(key_states, dim=-1)
+
+            value_states = [F.linear(hidden_states, value_slices[i]) for i in range(self.pretraining_tp)]
+            value_states = torch.cat(value_states, dim=-1)
+
+        else:
+            query_states = self.q_proj(hidden_states)
+            key_states = self.k_proj(hidden_states)
+            value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        # cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        # query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+        if past_key_value is not None:
+            # reuse k, v, self_attention
+            key_states = torch.cat([past_key_value[0], key_states], dim=2)
+            value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        max_length = max(query_states.shape[1], key_states.shape[1])
+        assert max_length <= self.sin.shape[0]
+        sin, cos = self.sin[:max_length], self.cos[:max_length]
+        # TODO: for inference, we can add emb kv into cache to avoid computation
+        query_states, key_states = apply_rotary_embedding(
+            query_states, key_states, cos, sin, decode=True if q_len == 1 else False, rotary_index=position_ids
+        )
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+
+        # repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3))
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            if self.training:
+                attention_mask = attention_mask.clone().detach()
+            attention_mask[:, :, :, 0] = 0
+            attn_weights = attn_weights + attention_mask
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.num_heads * self.head_dim)
+
+        if self.pretraining_tp > 1:
+            attn_output = attn_output.split(self.hidden_size // self.pretraining_tp, dim=2)
+            o_proj_slices = self.o_proj.weight.split(self.hidden_size // self.pretraining_tp, dim=1)
+            attn_output = sum([F.linear(attn_output[i], o_proj_slices[i]) for i in range(self.pretraining_tp)])
+        else:
+            attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class HFOpenMoeDecoderLayer(nn.Module):
+    def __init__(self, config: HFOpenMoeConfig, moe: bool):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.moe = moe
+        self.self_attn = HFOpenMoeAttention(config=config)
+        #         self.self_attn = LlamaAttention(config=config)  # TODO: introduce LLaMA Positional Encoding
+        self.input_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        if self.moe:
+            self.mlp = HFOpenMoeSparseMLP(config)
+            self.pre_extra_mlp_layernorm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+            self.extra_mlp = HFOpenMoeMLP(config)
+        else:
+            self.mlp = HFOpenMoeMLP(config)
+
+    def forward(
+            self,
+            hidden_states: torch.Tensor,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_value: Optional[Tuple[torch.Tensor]] = None,
+            output_attentions: Optional[bool] = False,
+            use_cache: Optional[bool] = False,
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, 1, tgt_len, src_len)` where padding elements are indicated by very large negative values.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        if self.moe:
+            residual = hidden_states
+            hidden_states = self.pre_extra_mlp_layernorm(hidden_states)
+            hidden_states = self.extra_mlp(hidden_states)
+            hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+LLAMA_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`HFOpenMoeConfig`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+    LLAMA_START_DOCSTRING,
+)
+class HFOpenMoePreTrainedModel(PreTrainedModel):
+    config_class = HFOpenMoeConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["HFOpenMoeDecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, HFOpenMoeModel):
+            module.gradient_checkpointing = value
+
+
+LLAMA_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `decoder_input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`tuple(tuple(torch.FloatTensor))`, *optional*, returned when `use_cache=True` is passed or when `config.use_cache=True`):
+            Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of shape
+            `(batch_size, num_heads, sequence_length, embed_size_per_head)`) and 2 additional tensors of shape
+            `(batch_size, num_heads, encoder_sequence_length, embed_size_per_head)`.
+
+            Contains pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used (see `past_key_values` input) to speed up sequential decoding.
+
+            If `past_key_values` are used, the user can optionally input only the last `decoder_input_ids` (those that
+            don't have their past key value states given to this model) of shape `(batch_size, 1)` instead of all
+            `decoder_input_ids` of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        use_cache (`bool`, *optional*):
+            If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding (see
+            `past_key_values`).
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare LLaMA Model outputting raw hidden-states without any specific head on top.",
+    LLAMA_START_DOCSTRING,
+)
+class HFOpenMoeModel(HFOpenMoePreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`LlamaDecoderLayer`]
+
+    Args:
+        config: HFOpenMoeConfig
+    """
+
+    def __init__(self, config: HFOpenMoeConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [
+                HFOpenMoeDecoderLayer(config, moe=True if (i + 1) % config.moe_layer_interval == 0 else False)
+                for i in range(config.num_hidden_layers)
+            ]
+        )
+        self.norm = LlamaRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(self, attention_mask, input_shape, inputs_embeds, past_key_values_length):
+        # create causal mask
+        # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape,
+                inputs_embeds.dtype,
+                device=inputs_embeds.device,
+                past_key_values_length=past_key_values_length,
+            )
+
+        if attention_mask is not None:
+            # [bsz, seq_len] -> [bsz, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]).to(
+                inputs_embeds.device
+            )
+            combined_attention_mask = (
+                expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+            )
+
+        return combined_attention_mask
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time")
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError("You have to specify either decoder_input_ids or decoder_inputs_embeds")
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+
+        if past_key_values is not None:
+            past_key_values_length = past_key_values[0][0].shape[2]
+            seq_length_with_past = seq_length_with_past + past_key_values_length
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length, seq_length + past_key_values_length, dtype=torch.long, device=device
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # embed positions
+        if attention_mask is None:
+            attention_mask = torch.ones(
+                (batch_size, seq_length_with_past), dtype=torch.bool, device=inputs_embeds.device
+            )
+        attention_mask = self._prepare_decoder_attention_mask(
+            attention_mask, (batch_size, seq_length), inputs_embeds, past_key_values_length
+        )
+
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = past_key_values[idx] if past_key_values is not None else None
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, output_attentions, None)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                    None,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(v for v in [hidden_states, next_cache, all_hidden_states, all_self_attns] if v is not None)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class HFOpenMoeForCausalLM(HFOpenMoePreTrainedModel):
+    # _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = HFOpenMoeModel(config)
+        self.pretraining_tp = config.pretraining_tp
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @add_start_docstrings_to_model_forward(LLAMA_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+            self,
+            input_ids: torch.LongTensor = None,
+            attention_mask: Optional[torch.Tensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[List[torch.FloatTensor]] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            labels: Optional[torch.LongTensor] = None,
+            use_cache: Optional[bool] = None,
+            output_attentions: Optional[bool] = None,
+            output_hidden_states: Optional[bool] = None,
+            return_dict: Optional[bool] = None,
+            chunk_head: Optional[bool] = True,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+                config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+                (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, LlamaForCausalLM
+
+        >>> model = LlamaForCausalLM.from_pretrained(PATH_TO_CONVERTED_WEIGHTS)
+        >>> tokenizer = AutoTokenizer.from_pretrained(PATH_TO_CONVERTED_TOKENIZER)
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+        # reset moe loss
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+
+        hidden_states = outputs[0]
+        if self.pretraining_tp > 1:
+            lm_head_slices = self.lm_head.weight.split(self.vocab_size // self.pretraining_tp, dim=0)
+            logits = [F.linear(hidden_states, lm_head_slices[i]) for i in range(self.pretraining_tp)]
+            logits = torch.cat(logits, dim=-1)
+
+        loss = None
+        # if no training, just do forward
+        if labels is None:
+            logits = self.lm_head(hidden_states)
+            logits = logits.float()
+        # the vocab size for openmoe is 30w+
+        # which causes great activation memory in training, up to 20G for one sequence
+        # so we use chunk and checkpoint to reduce memory
+        else:
+            if chunk_head == True:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        logits = module(inputs[0])
+                        logits = logits.float()
+                        # Shift so that tokens < n predict n
+                        shift_logits = logits[..., :-1, :].contiguous().float()
+                        shift_labels = inputs[1][..., 1:].contiguous()
+                        # Flatten the tokens
+                        loss = self._calculate_loss(shift_logits, shift_labels)
+                        return loss
+
+                    return custom_forward
+
+                for batch_idx in range(hidden_states.shape[0]):
+                    loss = loss + torch.utils.checkpoint.checkpoint(
+                        create_custom_forward(self.lm_head),
+                        hidden_states[batch_idx: batch_idx + 1, :],
+                        labels[batch_idx: batch_idx + 1, :],
+                    ) if loss is not None else torch.utils.checkpoint.checkpoint(
+                        create_custom_forward(self.lm_head),
+                        hidden_states[batch_idx: batch_idx + 1, :],
+                        labels[batch_idx: batch_idx + 1, :],
+                    )
+                logits = None
+            else:
+                logits = self.lm_head(hidden_states)
+                logits = logits.float()
+                # Shift so that tokens < n predict n
+                shift_logits = logits[..., :-1, :].contiguous()
+                shift_labels = labels[..., 1:].contiguous()
+                # Flatten the tokens
+                loss = self._calculate_loss(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+            self, input_ids, past_key_values=None, attention_mask=None, inputs_embeds=None, **kwargs
+    ):
+        if past_key_values:
+            input_ids = input_ids[:, -1:]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past),
+            )
+        return reordered_past
+
+    def _calculate_loss(self, logits: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:
+        """Compute cross entropy and entropy for log probs and targets.
+
+        Args:
+            logits: [batch, length, num_classes] float array.
+            targets: categorical targets [batch, length] int array.
+
+        Returns:
+            Tuple of scalar loss.
+        """
+        if len(logits.shape) != len(targets.shape) + 1:
+            raise ValueError(
+                "Incorrect shapes. Got shape %s logits and %s targets" % (str(logits.shape), str(targets.shape))
+            )
+        vocab_size = logits.shape[-1]
+        confidence = 1.0 - self.config.label_smoothing
+        low_confidence = (1.0 - confidence) / (vocab_size - 1)
+        normalizing_constant = -(
+                confidence * math.log(confidence) + (vocab_size - 1) * low_confidence * math.log(low_confidence + 1e-20)
+        )
+
+        # one hot
+        soft_targets = targets[..., None] == torch.arange(vocab_size, device=targets.device).reshape(
+            (1,) * len(targets.shape) + (-1,)
+        )
+        soft_targets = torch.where(
+            soft_targets, torch.full_like(soft_targets, confidence), torch.full_like(soft_targets, low_confidence)
+        )
+        soft_targets = soft_targets.to(torch.float32)
+
+        # cross entropy
+        total_loss = ZLossCrossEntropy.apply(logits, soft_targets, self.config.z_loss_factor)
+        total_loss = total_loss - normalizing_constant
+        total_loss = torch.mean(torch.sum(total_loss, dim=-1), dim=0)
+        return total_loss
+
+
+class ZLossCrossEntropy(torch.autograd.Function):
+    """Computes cross entropy loss with stable custom gradient.
+
+    Computes a stabilized-gradient version of:
+        -jnp.sum(targets * nn.log_softmax(logits), axis=-1)
+
+    If z_loss > 0, then an auxiliary loss equal to z_loss*log(z)^2
+    will be added to the cross entropy loss (z = softmax normalization constant).
+    The two uses of z_loss are:
+    1. To keep the logits from drifting too far from zero, which can cause
+        unacceptable roundoff errors in bfloat16.
+    2. To encourage the logits to be normalized log-probabilities.
+
+    Args:
+        logits: [batch, length, num_classes] float array.
+        targets: categorical one-hot targets [batch, length, num_classes] float
+        array.
+        z_loss: coefficient for auxilliary z-loss loss term.
+
+    Returns:
+        tuple with the total loss and the z_loss, both
+        float arrays with shape [batch, length].
+    """
+
+    @staticmethod
+    def forward(ctx, logits, targets, z_loss):
+        max_logit = torch.max(logits, dim=-1, keepdim=True)[0]
+        shifted = logits - max_logit
+        exp_shifted = torch.exp(shifted)
+        sum_exp = torch.sum(exp_shifted, axis=-1, keepdims=True)
+        sum_exp_log = torch.log(sum_exp)
+        log_softmax = shifted - sum_exp_log
+        loss = -torch.sum(targets * log_softmax, axis=-1)
+        # Add auxilliary z-loss term.
+        log_z = torch.squeeze(sum_exp_log + max_logit, axis=-1)
+        total_z_loss = z_loss * torch.square(log_z)
+        loss += total_z_loss
+        ctx.z_loss = z_loss
+        ctx.save_for_backward(logits, targets, exp_shifted, sum_exp, log_softmax, log_z)
+        return loss
+
+    @staticmethod
+    def backward(ctx, *grad_outputs):
+        assert len(grad_outputs) == 1
+        g = grad_outputs[0]
+        z_loss = ctx.z_loss
+        logits, targets, exp_shifted, sum_exp, log_softmax, log_z = ctx.saved_tensors
+        # z-loss term adds the (2 * z_loss * log_z) factor.
+        deriv = (1 + 2 * z_loss * log_z).unsqueeze(-1) * exp_shifted / sum_exp - targets
+        g_logits = g.unsqueeze(-1) * deriv
+        g_targets = -g.unsqueeze(-1) * log_softmax
+
+        return (
+            g_logits.to(logits.dtype),
+            g_targets.to(targets.dtype),
+            None,
+        )

tokenization_hf_openmoe.py

@@ -0,0 +1,24 @@
+from typing import List, Optional
+
+from transformers import T5Tokenizer
+
+
+class HFOpenMoeTokenizer(T5Tokenizer):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        self.padding_side = 'left'
+        self.add_bos_token = True
+        self.add_eos_token = False
+
+    def build_inputs_with_special_tokens(
+            self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
+    ) -> List[int]:
+        if self.add_eos_token:
+            token_ids_0 = self._add_eos_if_not_present(token_ids_0)
+        if self.add_bos_token:
+            token_ids_0 = [self.pad_token_id] + token_ids_0
+        if token_ids_1 is None:
+            return token_ids_0
+        else:
+            token_ids_1 = self._add_eos_if_not_present(token_ids_1)
+            return token_ids_0 + token_ids_1