config.json

{
  "architectures": [
    "Lfm2MllmForCausalLM"
  ],
  "block_auto_adjust_ff_dim": true,
  "block_ffn_dim_multiplier": 1.0,
  "block_multiple_of": 256,
  "bos_token_id": 1,
  "conv_L_cache": 3,
  "conv_bias": false,
  "eos_token_id": 2,
  "hidden_size": 2048,
  "image_aspect_ratio": "square",
  "image_end_token": "<im_end>",
  "image_grid_pinpoints": null,
  "image_patch_token": "<im_patch>",
  "image_start_token": "<im_start>",
  "image_token": "<image>",
  "image_token_id": 32000,
  "initializer_range": 0.02,
  "intermediate_size": 8192,
  "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"
  ],
  "max_position_embeddings": 128000,
  "mm_hidden_size": 2048,
  "mm_intermediate_size": 8192,
  "mm_layer_norm_eps": 1e-05,
  "mm_num_attention_heads": 32,
  "mm_num_hidden_layers": 2,
  "mm_use_im_patch_token": true,
  "mm_use_im_start_end": true,
  "model_type": "lfm2",
  "norm_eps": 1e-05,
  "num_attention_heads": 32,
  "num_hidden_layers": 24,
  "num_key_value_heads": 8,
  "pad_token_id": 0,
  "pretrain_mm_mlp_adapter": false,
  "rope_theta": 1000000.0,
  "torch_dtype": "float32",
  "transformers_version": "4.54.1",
  "tune_mm_mlp_adapter": false,
  "use_cache": true,
  "vision_config": {
    "architecture": "mobilenetv5_300m_enc",
    "do_pooling": false,
    "hidden_size": 2048,
    "initializer_range": 0.02,
    "label_names": [
      "LABEL_0",
      "LABEL_1"
    ],
    "model_args": null,
    "model_type": "gemma3n_vision",
    "num_classes": 2,
    "rms_norm_eps": 1e-06,
    "torch_dtype": "bfloat16",
    "vocab_offset": 262144,
    "vocab_size": 128
  },
  "vision_pretrained_path": null,
  "vision_projection_dim": 2048,
  "vision_soft_tokens_per_image": 256,
  "vocab_size": 32000
}