config.json

{
  "acoustic_tokenizer_chunk_size": 1440000,
  "acoustic_tokenizer_encoder_config": {
    "channels": 1,
    "depths": [
      3,
      3,
      3,
      3,
      3,
      3,
      8
    ],
    "downsampling_ratios": [
      2,
      2,
      4,
      5,
      5,
      8
    ],
    "dtype": "bfloat16",
    "ffn_expansion": 4,
    "hidden_act": "gelu",
    "hidden_size": 64,
    "initializer_range": 0.01,
    "kernel_size": 7,
    "layer_scale_init_value": 1e-06,
    "model_type": "vibevoice_acoustic_tokenizer_encoder",
    "num_filters": 32,
    "rms_norm_eps": 1e-05,
    "vae_std": 0.625,
    "weight_init_value": 0.01
  },
  "architectures": [
    "VibeVoiceAsrForConditionalGeneration"
  ],
  "audio_bos_token_id": 151646,
  "audio_eos_token_id": 151647,
  "audio_token_id": 151648,
  "dtype": "bfloat16",
  "model_type": "vibevoice_asr",
  "quantization_config": {
    "_load_in_4bit": true,
    "_load_in_8bit": false,
    "bnb_4bit_compute_dtype": "bfloat16",
    "bnb_4bit_quant_storage": "uint8",
    "bnb_4bit_quant_type": "nf4",
    "bnb_4bit_use_double_quant": true,
    "llm_int8_enable_fp32_cpu_offload": false,
    "llm_int8_has_fp16_weight": false,
    "llm_int8_skip_modules": [
      "acoustic_tokenizer_encoder",
      "semantic_tokenizer_encoder",
      "acoustic_projection",
      "semantic_projection",
      "lm_head"
    ],
    "llm_int8_threshold": 6.0,
    "load_in_4bit": true,
    "load_in_8bit": false,
    "quant_method": "bitsandbytes"
  },
  "semantic_tokenizer_encoder_config": {
    "channels": 1,
    "depths": [
      3,
      3,
      3,
      3,
      3,
      3,
      8
    ],
    "downsampling_ratios": [
      2,
      2,
      4,
      5,
      5,
      8
    ],
    "dtype": "bfloat16",
    "ffn_expansion": 4,
    "hidden_act": "gelu",
    "hidden_size": 128,
    "initializer_range": 0.01,
    "kernel_size": 7,
    "layer_scale_init_value": 1e-06,
    "model_type": "vibevoice_acoustic_tokenizer_encoder",
    "num_filters": 32,
    "rms_norm_eps": 1e-05,
    "vae_std": 0.625,
    "weight_init_value": 0.01
  },
  "text_config": {
    "attention_dropout": 0.0,
    "bos_token_id": null,
    "dtype": "bfloat16",
    "eos_token_id": null,
    "hidden_act": "silu",
    "hidden_size": 3584,
    "initializer_range": 0.02,
    "intermediate_size": 18944,
    "layer_types": [
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention",
      "full_attention"
    ],
    "max_position_embeddings": 131072,
    "max_window_layers": 28,
    "model_type": "qwen2",
    "num_attention_heads": 28,
    "num_hidden_layers": 28,
    "num_key_value_heads": 4,
    "pad_token_id": null,
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
      "rope_theta": 1000000.0,
      "rope_type": "default"
    },
    "sliding_window": null,
    "tie_word_embeddings": false,
    "use_cache": true,
    "use_mrope": false,
    "use_sliding_window": false,
    "vocab_size": 152064
  },
  "transformers_version": "5.3.0"
}