Spaces:

open-llm-leaderboard

/

open_llm_leaderboard

like 12k

Running on CPU Upgrade

Interesting! We'll look into it

Ya this would be nice, because the top models are getting replaced from time to time and it would be cool to see how long a model stays on it's specific spot for sometimes.

It could be like after the model name
modelorg/modelname ([⭐new]2024/04/18)
modelorg/modelname (2024/04/18)

What I was thinking is to have a column containing the date the model was added and an option to filter models added after a certain date. That would be useful to me. I might be able to help develop it but I really am not proficient with Gradio.

I already wrote a python script to grab the data including what I think is the date it was added and export it to csv...

At the moment we're in a bit of a rush but I think that's a good feature, we'll look into adding it :)

Let me add a similar suggestion: a column with the date when the model was added to the leaderboard.

Shouldn't be too hard to do, but would be very useful for following the leaderboard developments. Personally I'd be happy just to be able to sort the table by this column.

Edit: is the "date added to hub", suggested above, actually "date uploaded to HF" (as I understood it initially), or date added to leaderboard? Both would be useful.

Hi everyone!

We've added two new columns:

• Submission Date – when a model was submitted according to its request file
• Upload To Hub Date – when a model was uploaded to the hub

Please, check them out!

I didn't add a slider for the Upload To Hub Date column so as not to overcomplicate the system, but we'll think about how to filter that column when there will be more models on the Leaderboard

On a second look, submission date is not ideal if the goal is being able to see the latest additions to the leaderboard by sorting the appropriate column. For a model that just appeared in the leaderboard, submission date may be weeks or months in the past.

Hopefully we'll catch up on the queue and submission date will not be too far from the evaluation date!
It's also interesting as an indicator of community interest for a model :)

Since @alozowski solved this discussion, I'm closing it!