{
    "architectures": [
        "CohereForCausalLM"
    ],
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": 5,
    "eos_token_id": 6,
    "hidden_act": "silu",
    "hidden_size": 8192,
    "initializer_range": 0.02,
    "intermediate_size": 22528,
    "layer_norm_eps": 1e-05,
    "logit_scale": 0.0625,
    "max_position_embeddings": 8192,
    "model_max_length": 131072,
    "model_type": "cohere",
    "num_attention_heads": 64,
    "num_hidden_layers": 40,
    "num_key_value_heads": 64,
    "pad_token_id": 0,
    "pretraining_tp": 1,
    "quantization": {
        "group_size": 64,
        "bits": 4
    },
    "quantization_config": {
        "group_size": 64,
        "bits": 4
    },
    "rms_norm_eps": 1e-05,
    "rope_theta": 8000000.0,
    "torch_dtype": "bfloat16",
    "transformers_version": "4.38.2",
    "use_cache": true,
    "vocab_size": 256000
}