{
  "architectures": [
    "HYV4VLForConditionalGeneration"
  ],
  "dtype": "float32",
  "image_token_id": 120120,
  "model_type": "hy_v4_vl",
  "text_config": {
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": 120000,
    "eos_token_id": 120025,
    "hc_eps": 1e-06,
    "hc_magnitude": 2.0,
    "hc_mult": 2,
    "head_dim": 16,
    "hidden_act": "silu",
    "hidden_size": 64,
    "index_head_dim": 16,
    "index_n_heads": 2,
    "index_topk": 32,
    "indexer_types": [
      "full",
      "shared"
    ],
    "initializer_range": 0.006,
    "intermediate_size": 128,
    "kv_lora_rank": 16,
    "layer_types": [
      "indexed_attention",
      "indexed_attention"
    ],
    "learnable_sink_init": 0.0,
    "max_position_embeddings": 262144,
    "mlp_layer_types": [
      "dense",
      "sparse"
    ],
    "model_type": "hy_v4_vl_text",
    "moe_intermediate_size": 32,
    "n_group": 1,
    "n_routed_experts": 4,
    "n_shared_experts": 1,
    "norm_topk_prob": true,
    "num_attention_heads": 4,
    "num_experts_per_tok": 2,
    "num_hidden_layers": 2,
    "num_key_value_heads": 4,
    "output_router_logits": false,
    "pad_token_id": 120002,
    "q_lora_rank": 32,
    "qk_head_dim": 32,
    "qk_nope_head_dim": 16,
    "qk_rope_head_dim": 16,
    "rms_norm_eps": 1e-05,
    "rope_parameters": {
      "mrope_section": [
        4,
        2,
        2
      ],
      "rope_theta": 10000.0,
      "rope_type": "default"
    },
    "routed_scaling_factor": 2.827,
    "swiglu_limit": 10.0,
    "tie_word_embeddings": false,
    "topk_group": 1,
    "use_cache": true,
    "v_head_dim": 16,
    "vocab_size": 120832
  },
  "tie_word_embeddings": false,
  "transformers_version": "5.19.0.dev0",
  "video_end_token_id": 120123,
  "video_start_token_id": 120122,
  "video_token_id": 120124,
  "vision_config": {
    "attention_dropout": 0.0,
    "hidden_act": "gelu",
    "hidden_size": 32,
    "intermediate_size": 64,
    "interpolate_mode": "bilinear",
    "max_image_size": 2048,
    "model_type": "hy_v4_vl_vision",
    "num_attention_heads": 2,
    "num_channels": 3,
    "num_hidden_layers": 2,
    "num_key_value_heads": 2,
    "out_hidden_size": 64,
    "patch_size": 16,
    "rms_norm_eps": 1e-05,
    "spatial_merge_size": 2
  }
}