config.json

{
  "architectures": [
    "CSMModel"
  ],
  "audio_num_codebooks": 32,
  "audio_vocab_size": 2051,
  "backbone_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "LlamaForCausalLM"
    ],
    "head_dim": 64,
    "hidden_size": 2048,
    "intermediate_size": 8192,
    "model_type": "llama",
    "num_hidden_layers": 16,
    "num_key_value_heads": 8,
    "rms_norm_eps": 1e-05,
    "rope_scaling": {
      "factor": 32.0,
      "high_freq_factor": 4.0,
      "low_freq_factor": 1.0,
      "original_max_position_embeddings": 8192,
      "rope_type": "llama3",
      "type": "llama3"
    },
    "rope_theta": 500000,
    "vocab_size": 128256
  },
  "decoder_config": {
    "_attn_implementation_autoset": true,
    "architectures": [
      "LlamaForCausalLM"
    ],
    "hidden_size": 1024,
    "intermediate_size": 8192,
    "max_position_embeddings": 32,
    "model_type": "llama",
    "num_attention_heads": 8,
    "num_hidden_layers": 4,
    "num_key_value_heads": 2,
    "rms_norm_eps": 1e-05,
    "rope_scaling": {
      "factor": 32.0,
      "high_freq_factor": 4.0,
      "low_freq_factor": 1.0,
      "original_max_position_embeddings": 8192,
      "rope_type": "llama3",
      "type": "llama3"
    },
    "rope_theta": 500000,
    "vocab_size": 128256
  },
  "max_seq_len": 2048,
  "model_type": "csm",
  "text_vocab_size": 128256,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.49.0"
}