AnsenIO/dia_ttv
05
1{
2 "architectures": [
3 "DiaForConditionalGeneration"
4 ],
5 "bos_token_id": 1026,
6 "decoder_config": {
7 "cross_head_dim": 128,
8 "cross_hidden_size": 1024,
9 "cross_num_attention_heads": 16,
10 "cross_num_key_value_heads": 16,
11 "head_dim": 128,
12 "hidden_act": "silu",
13 "hidden_size": 2048,
14 "initializer_range": 0.02,
15 "intermediate_size": 8192,
16 "max_position_embeddings": 3072,
17 "model_type": "dia_decoder",
18 "norm_eps": 1e-05,
19 "num_attention_heads": 16,
20 "num_channels": 9,
21 "num_hidden_layers": 18,
22 "num_key_value_heads": 4,
23 "rope_scaling": null,
24 "rope_theta": 10000.0,
25 "vocab_size": 1028
26 },
27 "delay_pattern": [
28 0,
29 8,
30 9,
31 10,
32 11,
33 12,
34 13,
35 14,
36 15
37 ],
38 "encoder_config": {
39 "head_dim": 128,
40 "hidden_act": "silu",
41 "hidden_size": 1024,
42 "initializer_range": 0.02,
43 "intermediate_size": 4096,
44 "max_position_embeddings": 1024,
45 "model_type": "dia_encoder",
46 "norm_eps": 1e-05,
47 "num_attention_heads": 16,
48 "num_hidden_layers": 12,
49 "num_key_value_heads": 16,
50 "rope_scaling": null,
51 "rope_theta": 10000.0,
52 "vocab_size": 256
53 },
54 "eos_token_id": 1024,
55 "initializer_range": 0.02,
56 "is_encoder_decoder": true,
57 "model_type": "dia",
58 "norm_eps": 1e-05,
59 "pad_token_id": 1025,
60 "torch_dtype": "float32",
61 "transformers_version": "4.53.0.dev0"
62}