Upload model_config.yaml with huggingface_hub
Browse files- model_config.yaml +43 -0
model_config.yaml
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
add_qkv_bias: true
|
| 2 |
+
asr_adapter: llamamlp
|
| 3 |
+
attn_dropout: 0.0
|
| 4 |
+
bias: false
|
| 5 |
+
block_size: 2048
|
| 6 |
+
force_align: false
|
| 7 |
+
gelu_approximate: none
|
| 8 |
+
head_size: 64
|
| 9 |
+
hf_config:
|
| 10 |
+
name: Qwen2-0.5B
|
| 11 |
+
org: Qwen
|
| 12 |
+
intermediate_size: 4864
|
| 13 |
+
lm_head_bias: false
|
| 14 |
+
mlp_class_name: LLaMAMLP
|
| 15 |
+
n_embd: 896
|
| 16 |
+
n_expert: 0
|
| 17 |
+
n_expert_per_token: 0
|
| 18 |
+
n_head: 14
|
| 19 |
+
n_layer: 24
|
| 20 |
+
n_query_groups: 2
|
| 21 |
+
name: Qwen2-0.5B
|
| 22 |
+
norm_class_name: RMSNorm
|
| 23 |
+
norm_eps: 1.0e-06
|
| 24 |
+
padded_vocab_size: 181120
|
| 25 |
+
padding_multiple: 512
|
| 26 |
+
parallel_residual: false
|
| 27 |
+
pos_type: rope
|
| 28 |
+
post_adapter: false
|
| 29 |
+
post_adapter_layers: 6
|
| 30 |
+
prompt_vocab_size: null
|
| 31 |
+
rope_base: 1000000
|
| 32 |
+
rope_condense_ratio: 1
|
| 33 |
+
rotary_percentage: 1
|
| 34 |
+
scale_embeddings: false
|
| 35 |
+
shared_attention_norm: false
|
| 36 |
+
tie_word_embeddings: true
|
| 37 |
+
use_pretrain_phoneme_emb: false
|
| 38 |
+
vocab_size: 50254
|
| 39 |
+
text_vocab_size: 152000
|
| 40 |
+
cat_audio_vocab_size: 29120
|
| 41 |
+
audio_vocab_size: 4160
|
| 42 |
+
whisper_adapter_dim: 768
|
| 43 |
+
vision_adapter_dim: 512
|