config.json

{
    "architectures": [
        "CohereForCausalLM"
    ],
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": 5,
    "eos_token_id": 255001,
    "hidden_act": "silu",
    "hidden_size": 12288,
    "initializer_range": 0.02,
    "intermediate_size": 33792,
    "layer_norm_eps": 1e-05,
    "logit_scale": 0.8333333333333334,
    "max_position_embeddings": 131072,
    "model_type": "cohere",
    "num_attention_heads": 96,
    "num_hidden_layers": 64,
    "num_key_value_heads": 8,
    "pad_token_id": 0,
    "rope_theta": 8000000,
    "torch_dtype": "float16",
    "transformers_version": "4.44.0",
    "use_cache": true,
    "use_qk_norm": true,
    "vocab_size": 256000,
    "quantization_config": {
        "quant_method": "exl2",
        "version": "0.2.0",
        "bits": 3.0,
        "head_bits": 6,
        "calibration": {
            "rows": 115,
            "length": 2048,
            "dataset": "(default)"
        }
    }
}