Spaces:
Running
on
CPU Upgrade
Erronous filter on model size
Hello,
I wanted to bring to your attention that the filter for model sizes doesn't appear to work properly with the "~7b" filter. In fact, this filter also allows the 70b filter to pass through it. It seems like it might be a straightforward issue to address.
Thanks,
There seems to be three bugs here:
- A few 70b models are incorrectly categorized as 7b
- if you click multiple filter checkboxes without waiting for the first one to finish processing, the filters don’t work correctly. Aside from being a logic bug, a filter checkbox shouldn’t require any processing time.
To fix the processing time issue, the leaderboard shouldn’t be making network calls when data is filtered. Fetch all the LLM data on pageload, store it locally, and then query the local data instead. It looks like column sorting already works this way, which is a great update compared to before!
If you’re worried about people leaving the window open permanently and getting outdated data, you can also set a JavaScript timer to automatically refresh the page every 24 hours.
Hi, we are aware of this issue, model size is not computed properly for GPTQ models. We are currently working on a fix, closing this, and we will let you know when it is fixed !