Skip to content

Question about inference configuration and reproduction of results #46

Description

@liys0304

Thank you for sharing this great work! I am currently testing the model using the provided code, but I found that my inference results are different from those reported in the paper. I want to make sure I am using the correct configuration for inference. Specifically, when running the following command:
torchrun --nproc-per-node=8 eval/eval_videomme.py --data_path ./data/VideoMME --version qwen --model_path ./checkpoints/longvu_qwen
The final accuracy I obtain is only 34.04%, which is much lower than what is presented in the paper.
Could you clarify what specific settings should be used in config for inference? Are there any additional preprocessing or postprocessing steps that might affect the results? I would really appreciate any guidance on how to best reproduce the reported results, thanks for your help!

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions