{
    "model_id": "llava",
    "text_decoder": {
        "num_layers": 32,
        "hidden_size": 4096,
        "num_attention_heads": 32,
        "num_query_groups": 32,
        "ffn_hidden_size": 11008,
        "add_bias_linear": false,
        "bias_activation_fusion": false,
        "gated_linear_unit": true,
        "apply_query_key_layer_scaling":false,
        "layernorm_zero_centered_gamma": false,
        "max_position_embeddings": 4096,
        "bias_dropout_fusion":false,
        "apply_rope_fusion": false,
        "attention_softmax_in_fp32": true,
        "attention_dropout": 0.0,
        "hidden_dropout": 0.0,
        "bf16": true,
        "params_dtype": "bf16",
        "deallocate_pipeline_outputs": true,
        "persist_layer_norm": true,
        "activation_func": "silu",
        "normalization": "RMSNorm",
        "language_vocab_size": 32000,
        "language_max_sequence_length": 4096,
        "lm_position_embedding_type": "rope",
        "recompute_granularity": "full",
        "recompute_method": "block",
        "recompute_num_layers": 32,
        "freeze": true,
        "ckpt_path": "/<your_vicuna_weights_path>/converted_vicuna.pt"
    },
    "image_encoder": {
        "vision_encoder": {
            "model_id": "clip",
            "num_layers": 23,
            "hidden_size": 1024,
            "num_attention_heads": 16,
            "num_query_groups": 16,
            "ffn_hidden_size": 4096,
            "post_layer_norm": false,
            "add_bias_linear": true,
            "add_qkv_bias": true,
            "hidden_dropout": 0.0,
            "attention_dropout": 0.0,
            "bf16": true,
            "params_dtype": "bf16",
            "gated_linear_unit": false,
            "kv_channels": 64,
            "layernorm_zero_centered_gamma": false,
            "bias_activation_fusion": false,
            "bias_dropout_fusion": false,
            "attention_softmax_in_fp32": true,
            "normalization": "LayerNorm",
            "apply_rope_fusion": false,
            "activation_func": "quick_gelu",
            "device": "npu",
            "add_class_token": true,
            "class_token_len": 1,
            "patch_size": 14,
            "image_size": 336,
            "freeze": true,
            "ckpt_path": "/<your_clip_weights_path>/converted_clip.pt"
        },
        "vision_projector": {
            "model_id": "mlp",
            "num_layers": 2,
            "gated_linear_unit": false,
            "bias_activation_fusion": false,
            "add_bias_linear": true,
            "input_size": 1024,
            "hidden_size": 4096,
            "ffn_hidden_size": 4096,
            "activation_func": "gelu",
            "bf16": true,
            "params_dtype": "bf16",
            "freeze": false,
            "ckpt_path": null
        }
    },
    "text_encoder": null,
    "video_encoder": null
}