{
    "model_name": "Qwen3-VL-4B-Instruct",
    "model_type": "vlm-qwen3_vl",
    "vm_cfg": {
        "model_type": "qwen3_vl",
        "arch": "qwen3_vision_encoder",
        "image_size": [
            448,
            448
        ],
        "patch_size": 16,
        "cls_embed": false,
        "hidden_size": 1024,
        "intermediate_size": 4096,
        "num_attention_heads": 16,
        "num_hidden_layers": 24,
        "hidden_act": "gelu_pytorch_tanh",
        "layer_norm_eps": 1e-06,
        "spatial_merge_size": 2,
        "temporal_patch_size": 2,
        "window_size": 0,
        "num_position_embeddings": 2304,
        "fullatt_block_indexes": [],
        "deepstack_visual_indexes": [
            5,
            11,
            17
        ]
    },
    "mm_cfg": {
        "num_layers": 2,
        "hidden_act": "silu",
        "mm_tokens_per_image": 196,
        "proj_dim": 2560,
        "downsample_factor": 1
    },
    "lm_cfg": {
        "model_type": "qwen3_vl_text",
        "data_type": "bfloat16",
        "arch": "qwen",
        "gen": "3",
        "size": "4b",
        "token_cfg": {
            "vocab_size": 151936
        },
        "rope_cfg": {
            "rope_theta": 5000000,
            "rope_local_base_freq": 5000000,
            "rope_scaling": {
                "factor": 1.0,
                "low_freq_factor": 0,
                "high_freq_factor": 0,
                "original_max_position_embeddings": 0,
                "long_factor": null,
                "short_factor": null,
                "rope_type": "mrope",
                "mrope_section": [
                    24,
                    20,
                    20
                ],
                "mrope_interleaved": true
            }
        },
        "attn_cfg": {
            "num_attention_heads": 32,
            "num_key_value_heads": 8,
            "head_dim": 128,
            "swa_enable": false,
            "swa_ratio": 0,
            "sliding_window": 0,
            "attention_bias": false,
            "attention_dropout": 0.0,
            "query_pre_attn_scalar": 0
        },
        "mlp_cfg": {
            "intermediate_size": 9728,
            "act": "silu",
            "num_layers": 3,
            "mlp_bias": false
        },
        "hidden_size": 2560,
        "num_hidden_layers": 36,
        "max_position_embeddings": 2048,
        "rms_norm_eps": 1e-06,
        "rms_norm_unit_offset": false,
        "layer_norms": [
            "pre_attn",
            "post_attn",
            "qk_norm"
        ],
        "layer_types": null,
        "attn_logit_softcapping": null,
        "final_logit_softcapping": null,
        "lm_head_num_splits": 3,
        "lm_head_split_dim": 50656,
        "lora_cfg": null,
        "conv_L_cache": 3,
        "conv_bias": false
    },
    "pipeline_cfg": {
        "system_prompt": null,
        "chat_template": null,
        "max_num_tokens": 2048,
        "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
        ],
        "future_token_mask_size": 128,
        "return_logits": false,
        "use_strided_kv_cache": false,
        "enable_filter_sharing": false,
        "quantize_embeddings": false,
        "split_mlp": false
    },
    "language_model_name": "Qwen3-VL-4B-Instruct_language",
    "vision_model_name": "Qwen3-VL-4B-Instruct_vision"
}