Spaces:
Running
on
CPU Upgrade
Idea about model details
Hi, what do you think about adding some kind of score of efficiency for the models or just enable grouping by size
So its much easier to see how good the model in specific size performs.
Furthermore it would be pretty cool to retrieve the languages from model card, so it would be possible to search for models in other languages than english
I like the direction this is thinking in. Basically the leaderboard also becomes a nice dataset with metadata about the models that where evaluated as well.
I think getting the lang tags should be pretty simple as you already get the license etc
The design of grouping is more complex.
I had visual grouping in mind but while depeding on gradio maybe checkbox based filtering algorithm is faster to add instead ?
@clefourrier as hf staff maybe ? what do you think about it ?
Hi!
Thank you for your suggestions!
For the model efficiency, there is already an efficiency leaderboard available which should contain all information you need. I like the idea of grouping by size, though it would need more people to actually convert their weights to safetensors so we can be accurate - I've added it to our todos.
Adding supported languages would make more sense imo if the benchmarks we used covered more languages than English, as we don't want to overwhelm the leaderboard with too much information