devkit/vlm_config.json

{
    "model_name": "LFM2-VL-3B",
    "model_type": "vlm-lfm2-vl",
    "vm_cfg": {
        "model_type": "siglip2_vision_model",
        "arch": "siglip2",
        "image_size": 512,
        "patch_size": 16,
        "cls_embed": false,
        "hidden_size": 1152,
        "intermediate_size": 4304,
        "num_attention_heads": 16,
        "num_hidden_layers": 27,
        "hidden_act": "gelu_pytorch_tanh",
        "layer_norm_eps": 1e-06,
        "spatial_merge_size": 0,
        "temporal_patch_size": 0,
        "window_size": 0,
        "num_position_embeddings": 0,
        "fullatt_block_indexes": [],
        "deepstack_visual_indexes": []
    },
    "mm_cfg": {
        "num_layers": 2,
        "hidden_act": "gelu",
        "mm_tokens_per_image": 256,
        "proj_dim": 2048,
        "downsample_factor": 2
    },
    "lm_cfg": {
        "model_type": "lfm2",
        "data_type": "bfloat16",
        "arch": "lfm",
        "gen": "2",
        "size": "3b",
        "token_cfg": {
            "vocab_size": 65536
        },
        "rope_cfg": {
            "rope_theta": 1000000.0,
            "rope_local_base_freq": 1000000,
            "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": "default",
                "mrope_section": null,
                "mrope_interleaved": false
            }
        },
        "attn_cfg": {
            "num_attention_heads": 32,
            "num_key_value_heads": 8,
            "head_dim": 64,
            "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": 10752,
            "act": "silu",
            "num_layers": 3,
            "mlp_bias": false
        },
        "hidden_size": 2048,
        "num_hidden_layers": 30,
        "max_position_embeddings": 2048,
        "rms_norm_eps": 1e-05,
        "rms_norm_unit_offset": false,
        "layer_norms": [
            "pre_attn",
            "post_attn"
        ],
        "layer_types": [
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv",
            "full_attention",
            "conv",
            "conv"
        ],
        "attn_logit_softcapping": null,
        "final_logit_softcapping": null,
        "lm_head_num_splits": 1,
        "lm_head_split_dim": 65536,
        "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": true
    },
    "language_model_name": "LFM2-VL-3B_language",
    "vision_model_name": "LFM2-VL-3B_vision"
}