devkit/vlm_config.json

{
    "model_name": "models--meta-llama--Llama-3.1-8B-Instruct",
    "model_type": "llm-llama",
    "vm_cfg": null,
    "mm_cfg": null,
    "lm_cfg": {
        "model_type": "llama",
        "data_type": "bfloat16",
        "arch": "llama",
        "token_cfg": {
            "vocab_size": 128256
        },
        "rope_cfg": {
            "rope_theta": 500000.0,
            "rope_local_base_freq": 500000.0,
            "rope_dimension_count": 128,
            "rope_scaling": {
                "factor": 8.0,
                "low_freq_factor": 1.0,
                "high_freq_factor": 4.0,
                "original_max_position_embeddings": 8192,
                "long_factor": null,
                "short_factor": null,
                "rope_type": "llama3",
                "mrope_section": null,
                "mrope_interleaved": false
            }
        },
        "attn_cfg": {
            "num_attention_heads": 32,
            "num_key_value_heads": 8,
            "head_dim": 128,
            "swa_enable": false,
            "sliding_window": 0,
            "attention_bias": false,
            "attention_dropout": 0.0,
            "query_pre_attn_scalar": 0
        },
        "mlp_cfg": {
            "intermediate_size": 14336,
            "act": "silu",
            "num_layers": 3,
            "mlp_bias": false
        },
        "hidden_size": 4096,
        "num_hidden_layers": 32,
        "max_position_embeddings": 2048,
        "rms_norm_eps": 1e-05,
        "rms_norm_unit_offset": false,
        "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",
            "full_attention",
            "full_attention",
            "full_attention",
            "full_attention"
        ],
        "attn_logit_softcapping": null,
        "final_logit_softcapping": null,
        "lm_head_num_splits": 2,
        "lm_head_split_dim": 64128,
        "conv_L_cache": 3,
        "conv_bias": false,
        "lora_cfg": null
    },
    "pipeline_cfg": {
        "system_prompt": null,
        "chat_template": null,
        "max_num_tokens": 4096,
        "input_token_group_size": 128,
        "input_token_group_offsets": [
            0,
            128,
            256,
            384,
            512,
            640,
            768,
            896,
            1024,
            1152,
            1280,
            1408,
            1536,
            1664,
            1792,
            1920,
            2048,
            2176,
            2304,
            2432,
            2560,
            2688,
            2816,
            2944,
            3072,
            3200,
            3328,
            3456,
            3584,
            3712,
            3840,
            3968
        ],
        "future_token_mask_size": 128,
        "return_logits": false,
        "use_strided_kv_cache": false,
        "enable_filter_sharing": false,
        "quantize_embeddings": false,
        "split_mlp": true
    },
    "language_model_name": "models--meta-llama--Llama-3.1-8B-Instruct_language"
}