{
  "architectures": [
    "GenLIPModel"
  ],
  "attention_dropout": 0.0,
  "drop_path_rate": 0.1,
  "gated_attention": true,
  "hidden_act": "silu",
  "hidden_size": 1024,
  "initializer_factor": 1.0,
  "initializer_range": 0.02,
  "intermediate_size": 2752,
  "layer_norm_eps": 1e-06,
  "llm_decoder": "Qwen/Qwen3-0.6B",
  "lm_head_hdsz": 1024,
  "ls_init_value": 0.1,
  "max_position_embeddings": 16384,
  "max_window_layers": 80,
  "model_type": "genlip",
  "mrope_interleaved": true,
  "num_attention_heads": 16,
  "num_channels": 3,
  "num_hidden_layers": 24,
  "num_key_value_heads": 16,
  "num_patches": 196,
  "out_hidden_size": 1152,
  "patch_size": 16,
  "post_init_method": null,
  "rms_norm_eps": 1e-05,
  "rope_scaling": {
    "mrope_section": [
      12,
      10,
      10
    ],
    "rope_type": "default",
    "type": "default"
  },
  "rope_theta": 10000.0,
  "sliding_window": 4096,
  "spatial_merge_size": 1,
  "temporal_patch_size": 1,
  "text_embed_dim": 1024,
  "tie_word_embeddings": false,
  "tokens_per_second": 2,
  "torch_dtype": "bfloat16",
  "transformers_version": "4.53.0",
  "use_cache": true,
  "use_liger_kernel": true,
  "use_llm_head": false,
  "use_sliding_window": false,
  "use_swiglu_ffn": true,
  "vision_end_token_id": 151653,
  "vision_start_token_id": 151652,
  "vocab_size": 151936
}