pretraiend model in DiffusionDPO/sift_dpo_data.py

[Darius-H]

UnifiedReward/DiffusionDPO/sift_dpo_data.py

Line 27 in 1669fa1

pretrained = "/path/to/reward_model"

What should I write to replace this line? I tried pretrained = "CodeGoat24/UnifiedReward-qwen-7b" but in LLava code, the vision_tower = model.get_vision_tower() is None. The error is below

  File "./projects/UnifiedReward/VideoDPO/sift_dpo_data.py", line 201, in <module>
    tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, device_map=device_map)  # Add any other thing you want to pass in llava_model_args
                                                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "./projects/UnifiedReward/VideoDPO/LLaVA/llava/model/builder.py", line 156, in load_pretrained_model
    if not vision_tower.is_loaded:
           ^^^^^^^^^^^^^^^^^^^^^^
AttributeError: 'NoneType' object has no attribute 'is_loaded'

[CodeGoat24]

Hi!

The current code loads a LLaVA-based UnifiedReward model and runs inference.

If you want to use the Qwen version instead, you just need to replace the model loading and inference parts.

[pokameng]

Hi!

The current code loads a LLaVA-based UnifiedReward model and runs inference.

If you want to use the Qwen version instead, you just need to replace the model loading and inference parts.

how to replace？can you give some guidance?