Update config.json
Browse files- config.json +39 -83
config.json
CHANGED
@@ -1,103 +1,59 @@
|
|
1 |
{
|
|
|
|
|
2 |
"auto_map": {
|
3 |
"AutoModelForCausalLM": "modeling_audio_only_thinker.AudioOnlyThinker"
|
4 |
},
|
5 |
-
"architectures": [
|
6 |
-
"AudioOnlyThinker"
|
7 |
-
],
|
8 |
-
"audio_config": {
|
9 |
-
"_attn_implementation_autoset": true,
|
10 |
-
"activation_dropout": 0.0,
|
11 |
-
"activation_function": "gelu",
|
12 |
-
"attention_dropout": 0.0,
|
13 |
-
"d_model": 1280,
|
14 |
-
"dropout": 0.0,
|
15 |
-
"encoder_attention_heads": 20,
|
16 |
-
"encoder_ffn_dim": 5120,
|
17 |
-
"encoder_layerdrop": 0.0,
|
18 |
-
"encoder_layers": 32,
|
19 |
-
"init_std": 0.02,
|
20 |
-
"max_source_positions": 1500,
|
21 |
-
"model_type": "qwen2_5_omni_audio_encoder",
|
22 |
-
"n_window": 100,
|
23 |
-
"num_hidden_layers": 32,
|
24 |
-
"num_mel_bins": 128,
|
25 |
-
"output_dim": 3584,
|
26 |
-
"scale_embedding": false
|
27 |
-
},
|
28 |
-
"audio_end_token_id": 151648,
|
29 |
-
"audio_start_token_id": 151647,
|
30 |
"audio_token_index": 151646,
|
|
|
|
|
31 |
"bos_token_id": 151644,
|
32 |
"eos_token_id": 151645,
|
33 |
-
"ignore_index": -100,
|
34 |
-
"image_token_index": 151655,
|
35 |
-
"init_std": 0.02,
|
36 |
-
"model_type": "qwen2_5_omni_thinker",
|
37 |
"pad_token_id": 151643,
|
38 |
-
"
|
39 |
-
"
|
|
|
40 |
"text_config": {
|
41 |
-
"
|
42 |
-
"
|
43 |
"hidden_size": 3584,
|
44 |
-
"init_std": 0.02,
|
45 |
"intermediate_size": 18944,
|
46 |
-
"max_position_embeddings": 32768,
|
47 |
-
"max_window_layers": 28,
|
48 |
-
"model_type": "qwen2_5_omni_text",
|
49 |
"num_attention_heads": 28,
|
50 |
"num_hidden_layers": 28,
|
51 |
"num_key_value_heads": 4,
|
52 |
-
"
|
|
|
|
|
|
|
53 |
"rope_scaling": {
|
54 |
-
"mrope_section": [
|
55 |
-
|
56 |
-
|
57 |
-
24
|
58 |
-
],
|
59 |
-
"rope_type": "default",
|
60 |
-
"type": "default"
|
61 |
},
|
62 |
-
"rope_theta": 1000000.0,
|
63 |
-
"sliding_window": 32768,
|
64 |
-
"tie_word_embeddings": false,
|
65 |
"use_cache": true,
|
66 |
"use_sliding_window": false,
|
67 |
-
"
|
68 |
-
},
|
69 |
-
"tie_word_embeddings": false,
|
70 |
-
"torch_dtype": "bfloat16",
|
71 |
-
"transformers_version": "4.50.0.dev0",
|
72 |
-
"user_token_id": 872,
|
73 |
-
"video_token_index": 151656,
|
74 |
-
"vision_config": {
|
75 |
-
"_attn_implementation_autoset": true,
|
76 |
-
"depth": 32,
|
77 |
-
"embed_dim": 1280,
|
78 |
-
"fullatt_block_indexes": [
|
79 |
-
7,
|
80 |
-
15,
|
81 |
-
23,
|
82 |
-
31
|
83 |
-
],
|
84 |
-
"hidden_act": "silu",
|
85 |
-
"hidden_size": 1280,
|
86 |
-
"in_channels": 3,
|
87 |
-
"in_chans": 3,
|
88 |
"init_std": 0.02,
|
89 |
-
"
|
90 |
-
"
|
91 |
-
"
|
92 |
-
"out_hidden_size": 3584,
|
93 |
-
"patch_size": 14,
|
94 |
-
"spatial_merge_size": 2,
|
95 |
-
"spatial_patch_size": 14,
|
96 |
-
"temporal_patch_size": 2,
|
97 |
-
"tokens_per_second": 25,
|
98 |
-
"window_size": 112
|
99 |
},
|
100 |
-
"
|
101 |
-
|
102 |
-
|
103 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
{
|
2 |
+
"model_type": "audio-only-thinker",
|
3 |
+
"architectures": ["AudioOnlyThinker"],
|
4 |
"auto_map": {
|
5 |
"AutoModelForCausalLM": "modeling_audio_only_thinker.AudioOnlyThinker"
|
6 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
"audio_token_index": 151646,
|
8 |
+
"audio_start_token_id": 151647,
|
9 |
+
"audio_end_token_id": 151648,
|
10 |
"bos_token_id": 151644,
|
11 |
"eos_token_id": 151645,
|
|
|
|
|
|
|
|
|
12 |
"pad_token_id": 151643,
|
13 |
+
"tie_word_embeddings": false,
|
14 |
+
"torch_dtype": "bfloat16",
|
15 |
+
"transformers_version": "4.50.0.dev0",
|
16 |
"text_config": {
|
17 |
+
"model_type": "qwen2_5_omni_text",
|
18 |
+
"vocab_size": 152064,
|
19 |
"hidden_size": 3584,
|
|
|
20 |
"intermediate_size": 18944,
|
|
|
|
|
|
|
21 |
"num_attention_heads": 28,
|
22 |
"num_hidden_layers": 28,
|
23 |
"num_key_value_heads": 4,
|
24 |
+
"max_position_embeddings": 32768,
|
25 |
+
"sliding_window": 32768,
|
26 |
+
"max_window_layers": 28,
|
27 |
+
"rope_theta": 1000000.0,
|
28 |
"rope_scaling": {
|
29 |
+
"mrope_section": [16, 24, 24],
|
30 |
+
"type": "default",
|
31 |
+
"rope_type": "default"
|
|
|
|
|
|
|
|
|
32 |
},
|
|
|
|
|
|
|
33 |
"use_cache": true,
|
34 |
"use_sliding_window": false,
|
35 |
+
"tie_word_embeddings": false,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
"init_std": 0.02,
|
37 |
+
"rms_norm_eps": 1e-06,
|
38 |
+
"hidden_act": "silu",
|
39 |
+
"attention_dropout": 0.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
},
|
41 |
+
"audio_config": {
|
42 |
+
"model_type": "qwen2_5_omni_audio_encoder",
|
43 |
+
"d_model": 1280,
|
44 |
+
"num_hidden_layers": 32,
|
45 |
+
"encoder_attention_heads": 20,
|
46 |
+
"encoder_ffn_dim": 5120,
|
47 |
+
"dropout": 0.0,
|
48 |
+
"activation_function": "gelu",
|
49 |
+
"num_mel_bins": 128,
|
50 |
+
"output_dim": 3584,
|
51 |
+
"max_source_positions": 1500,
|
52 |
+
"n_window": 100,
|
53 |
+
"scale_embedding": false,
|
54 |
+
"init_std": 0.02,
|
55 |
+
"activation_dropout": 0.0,
|
56 |
+
"attention_dropout": 0.0,
|
57 |
+
"_attn_implementation_autoset": true
|
58 |
+
}
|
59 |
+
}
|