Unable to reproduce Qwen2.5-VL-7B-Instruct

Open bollossom opened this issue 10 months ago • 1 comments

How nice a work VLMEvalKit it is!!!

I tested Qwen2.5-VL-7B-Instruct on mmstar and it was only 61.8, but it was 63.9 in the text. The environment version is as follows

transformers==4.49.0 torch==2.6.0 flash_attn==2.7.4.post1

Feb 27 '25 12:02 bollossom

Please use pip install git+https://github.com/huggingface/transformers accelerate to build transformers from source. We use latest transformers and torch=2.5.1.

Feb 28 '25 02:02 PhoenixZ810