Leaderboard: Add more explainers

[Muennighoff]

Similar to https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/ could be via hovering; Things to explain imo:

• Max Tokens
• Embedding Dimension
• Number of Parameters

Let's make the LB maximally accessible

[KennethEnevoldsen]

I completely agree. We have already looked into this but haven't found it within gradio. We could add an FAQ box at the bottom, but I don't think it is needed for these features (we e.g. have it for zero-shot)

[KennethEnevoldsen]

will close this, but if there is potential solutions please do re-open.

[Muennighoff]

How about an accordion sth like the below next to zero-shot, borda?

More information

Number of Parameters: This is the total number of parameters in the model including embedding parameters. A higher value means the model requires more CPU/GPU memory to run; thus, less is generally desirable.
Embedding Dimension: This is the vector dimension of the embeddings that the model produces. When saving embeddings to disk, a higher dimension will require more space, thus less is usually desirable.
Max tokens: This refers to how many tokens (=word pieces) the model can process. Generally, a larger value is desirable.

[KennethEnevoldsen]

Would be happy with that - potentially rephrased slightly?

What do the columns mean?

• Number of Parameters: This is the total number of parameters in the model including embedding parameters. A higher value means the model requires more CPU/GPU memory to run; thus, less is generally desirable.
• Embedding Dimension: This is the vector dimension of the embeddings that the model produces. When saving embeddings to disk, a higher dimension will require more space, thus less is usually desirable.
• Max tokens: This refers to how many tokens (=word pieces) the model can process. Generally, a larger value is desirable.
• Zero-shot: ...

[Muennighoff]

Yeah that looks great!!