ml-energy / leaderboard Goto Github PK

benchmark, batch = 2 of a100 are added and upload to branch "a100_batch", but the response_length of batch = 2 is still shorter than batch = 1, though attention mask is added

it's better to handle batching and the edge cases by subclassing torch.utils.data.Dataset, but not manually writing code for.

Coding has become an important task for LLMs. We have good publicly available models too: CodeLlama, StarCoder, SantaCoder, Salesforce CodeT5 & CodeGen, etc. There's also a nice evaluation planforms like EvalPlus.

This chart goes against the principles of data visualization:

Recommendation to improve:

1. Remove all static texts around circles
2. Put the texts as tooltips so they pop up only when user hovers over them (I now see that the tooltips are present already)

You can use my code base to find out how to code tooltips with Plotly/Dash
https://github.com/pnlbwh/freesurfer-analysis/blob/7e00327599f4cb6e11561ca413b29b93f88f6a23/scripts/show-stats-table.py#L106-L114

Diffusion models for image generation is a workload that many people would be interested in.