{
  "_name_or_path": "RynnEC-2B-stage2",
  "architectures": [
    "RynnecQwen2ForCausalLM"
  ],
  "attention_dropout": 0.0,
  "bce_loss_weight": 2.0,
  "bos_token_id": 151643,
  "dice_loss_weight": 0.5,
  "eos_token_id": 151645,
  "has_mask": false,
  "hidden_act": "silu",
  "hidden_size": 1536,
  "image_aspect_ratio": "qwen2vl",
  "image_size": -1,
  "image_token_index": 151665,
  "image_token_length": 1,
  "initializer_range": 0.02,
  "intermediate_size": 8960,
  "is_alignment": false,
  "llm_lr": 4e-05,
  "loss_reduction_scope": "batch",
  "mask_decoder_model": "./checkpoints/sam2_hiera_large.pt",
  "max_frames": 180,
  "max_position_embeddings": 32768,
  "max_window_layers": 21,
  "mm_hidden_size": 1152,
  "mm_mask_decoder": true,
  "mm_max_length": 8192,
  "mm_projector_lr": 1e-05,
  "mm_projector_type": "mlp2x_gelu",
  "mm_vision_encoder": "DAMO-NLP-SG/VL3-SigLIP-NaViT",
  "mm_vision_select_feature": "patch",
  "mm_vision_select_layer": -2,
  "model_type": "rynnec_qwen2",
  "num_attention_heads": 12,
  "num_frames": 8,
  "num_hidden_layers": 28,
  "num_key_value_heads": 2,
  "out_dim": 256,
  "region_encoder_lr": 1e-05,
  "region_token_index": 151668,
  "rms_norm_eps": 1e-06,
  "rope_scaling": null,
  "rope_theta": 1000000.0,
  "sam_decoder_lr": 4e-05,
  "sam_encoder_lr": null,
  "seg_token_index": 151669,
  "sliding_window": null,
  "spatial_merge_size": 2,
  "tie_word_embeddings": true,
  "tokenizer_model_max_length": 16384,
  "tokenizer_padding_side": "right",
  "torch_dtype": "bfloat16",
  "training": true,
  "transformers_version": "4.46.3",
  "use_cache": false,
  "use_mm_proj": true,
  "use_sliding_window": false,
  "use_token_compression": true,
  "vision_encoder_lr": 0.0,
  "vocab_size": 151672
}