Yang Shi

icon indicating a website link https://dogneversleep.github.io icon indicating an email [email protected]

icon company Peking University icon location Beijing, China icon location 👨🏻‍🎓PhD student at Peking University | 🌠Interested in LLM / MLLM / AI Agent | 🐯 Life's a struggle.

Results 1 issues of Yang Shi

[Question] Cannot reproduce LLaVA-1.5-7b performance on MME with lmms-eval

5 comment

### Question Hi there. We cannot reproduce MME results following [finetune.sh](https://github.com/haotian-liu/LLaVA/blob/main/scripts/v1_5/finetune.sh) on 665k instruction tuning dataset. We followed all the training settings(pretrain+sft), but get 1497/289 on MME with lmms-eval.