config.json

{
  "architectures": [
    "DiaForConditionalGeneration"
  ],
  "bos_token_id": 1026,
  "decoder_config": {
    "cross_head_dim": 128,
    "cross_hidden_size": 1024,
    "cross_num_attention_heads": 16,
    "cross_num_key_value_heads": 16,
    "head_dim": 128,
    "hidden_act": "silu",
    "hidden_size": 2048,
    "initializer_range": 0.02,
    "intermediate_size": 8192,
    "max_position_embeddings": 3072,
    "model_type": "dia_decoder",
    "norm_eps": 1e-05,
    "num_attention_heads": 16,
    "num_channels": 9,
    "num_hidden_layers": 18,
    "num_key_value_heads": 4,
    "rope_scaling": null,
    "rope_theta": 10000.0,
    "vocab_size": 1028
  },
  "delay_pattern": [
    0,
    8,
    9,
    10,
    11,
    12,
    13,
    14,
    15
  ],
  "encoder_config": {
    "head_dim": 128,
    "hidden_act": "silu",
    "hidden_size": 1024,
    "initializer_range": 0.02,
    "intermediate_size": 4096,
    "max_position_embeddings": 1024,
    "model_type": "dia_encoder",
    "norm_eps": 1e-05,
    "num_attention_heads": 16,
    "num_hidden_layers": 12,
    "num_key_value_heads": 16,
    "rope_scaling": null,
    "rope_theta": 10000.0,
    "vocab_size": 256
  },
  "eos_token_id": 1024,
  "initializer_range": 0.02,
  "is_encoder_decoder": true,
  "model_type": "dia",
  "norm_eps": 1e-05,
  "pad_token_id": 1025,
  "torch_dtype": "float32",
  "transformers_version": "4.53.0.dev0"
}