Add model config
Browse files- config.json +45 -0
config.json
ADDED
|
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_qkv_bias": true,
|
| 3 |
+
"asr_adapter": "llamamlp",
|
| 4 |
+
"attn_dropout": 0,
|
| 5 |
+
"bias": false,
|
| 6 |
+
"block_size": 2048,
|
| 7 |
+
"force_align": false,
|
| 8 |
+
"gelu_approximate": "none",
|
| 9 |
+
"head_size": 64,
|
| 10 |
+
"hf_config": {
|
| 11 |
+
"name": "Qwen2-0.5B",
|
| 12 |
+
"org": "Qwen"
|
| 13 |
+
},
|
| 14 |
+
"intermediate_size": 4864,
|
| 15 |
+
"lm_head_bias": false,
|
| 16 |
+
"mlp_class_name": "LLaMAMLP",
|
| 17 |
+
"n_embd": 896,
|
| 18 |
+
"n_expert": 0,
|
| 19 |
+
"n_expert_per_token": 0,
|
| 20 |
+
"n_head": 14,
|
| 21 |
+
"n_layer": 24,
|
| 22 |
+
"n_query_groups": 2,
|
| 23 |
+
"name": "Qwen2-0.5B",
|
| 24 |
+
"norm_class_name": "RMSNorm",
|
| 25 |
+
"norm_eps": 0.000001,
|
| 26 |
+
"padded_vocab_size": 181120,
|
| 27 |
+
"padding_multiple": 512,
|
| 28 |
+
"parallel_residual": false,
|
| 29 |
+
"pos_type": "rope",
|
| 30 |
+
"post_adapter": false,
|
| 31 |
+
"post_adapter_layers": 6,
|
| 32 |
+
"prompt_vocab_size": null,
|
| 33 |
+
"rope_base": 1000000,
|
| 34 |
+
"rope_condense_ratio": 1,
|
| 35 |
+
"rotary_percentage": 1,
|
| 36 |
+
"scale_embeddings": false,
|
| 37 |
+
"shared_attention_norm": false,
|
| 38 |
+
"tie_word_embeddings": true,
|
| 39 |
+
"use_pretrain_phoneme_emb": false,
|
| 40 |
+
"vocab_size": 50254,
|
| 41 |
+
"text_vocab_size": 152000,
|
| 42 |
+
"cat_audio_vocab_size": 29120,
|
| 43 |
+
"audio_vocab_size": 4160,
|
| 44 |
+
"whisper_adapter_dim": 768
|
| 45 |
+
}
|