Spaces:

mteb

/

leaderboard

like 2.78k

Running on CPU Upgrade

Hello!

I'll defer to @Muennighoff for the final call as he knows best, but I think that we indeed accept models this way.
We only require that a model is usable in some way, whether that's via open weights of an API does not matter.
However, for API models it's more common to make a pull request to https://huggingface.co/datasets/mteb/results/tree/main/results to add a piccolo-large-zh-v2 folder with results. You can then make another pull request on https://huggingface.co/spaces/mteb/leaderboard/blob/main/app.py#L336 to add your model to that list of external models. You can then also add a link of your choice here and set the output dimensionality here. Especially that last one cannot be gathered if you upload your model scores in the way that you've currently done.

However, in the meantime I've restarted the leaderboard, congratulations!

• Tom Aarsen

Thanks for your suggestion !
But recently due to some reasons, the company's blog page cannot be updated. I will keep the content of this huggingface page for the next few days to give an introduction to this model.
Then I will try to use your method to pull a request of this model !.

@tomaarsen Hello, for the piccolo-v2 model：https://huggingface.co/sensenova/piccolo-large-zh-v2, I have uploaded the weights and parameter configs, but why are the parameters (dimension, model size) in the leaderboard still displayed as empty ?

Hello @Jinkin ,

I've restarted the leaderboard, and now the data shows correctly! Congratulations on your release - I think your multi-task approach is excellent. The MRL performance is also excellent, and I appreciate that you filtered to avoid CMTEB overlap!

• Tom Aarsen

Thanks , I see it !