The model performance actually not very good

ByungKwanLee / Meteor

[NeurIPS 2024] Official PyTorch implementation code for realizing the technical part of Mamba-based traversal of rationale (Meteor) to improve performance of numerous vision language performances for diverse capabilities.

MIT License

102 stars 4 forks source link

The model performance actually not very good #5

Open lucasjinreal opened 5 months ago

lucasjinreal commented 5 months ago

It's far worse than MiniCPM-LLama3 and Bunny-LLama3, even though your metric claims better than them

ByungKwanLee commented 5 months ago

Can you share your image and text in english please

Note that I did not implement history-saved conversation in demo