Spaces:
Running
on
CPU Upgrade
LLM Explorer: navigate thru 16,000 LLMs and explore their internals with ease
Hello Everyone,
I might be doing something wrong, but I've been experiencing some issues with the search box on HuggingFace, often receiving irrelevant results when attempting to explore different models. As a solution, I decided to create a comprehensive table to better facilitate these searches.
My creation provides filters by type such as uncensored, quantized, and supported languages. I'm pleased to say the database currently houses over 16,000 models. They are conveniently presented in a table format with flexible column filtering capabilities.
You can access the tool via this link: https://llm.extractum.io
Your feedback and bug reports would be greatly appreciated as it will assist in further refining the system.
I Love the idea! And thank you for all the work that you have done so far!
I have a few Suggestions & Questions for you.
Suggestions for your actual Data:
Looking at your current Database table sorted by "HF Rank" Shows falcon-40b-instruct as #1, (Shown In the Image Below)
While on The current Hugging Face Leaderboard, It has "llama-30b-instruct-2048" as the #1. (Shown In the Image Below)
So what this tells me is that your database is not fetching new data as it is posted from the leaderboard.
In order for your database to be considered credible and for it to be used as a helpful tool for the community, you are going to have to have it fetch and update the new data as it is posted.
Obviously, you could just do it manually and check it every 24 or 12 hours and update it accordingly, however, that would be a pain for you to have to do that, and it would also make your data less credible since some new data updates could sometimes be missed.
So you instead would have to set up a "data fetching" method where your table will auto-update every X hours. (X = Number of Hours) *the more frequent the better.Suggestions for your frontend/UI:
Overall Your UI isn't too bad. I think you are on the right track all things considered. However, one thing that I found annoying that you could improve right away was the way you style your Column Buttons. To be specific, whether or not the column is Enabled, or Disabled.
To help better explain what I am talking about, I have provided 2 images.
On the "Enabled" image (Shown In the Image Below), all of those buttons should be styled where they appear activated.
On The Disabled Image (Shown In the Image Below), I like that the remaining columns slid over to the left to fill the space of the column that is now disabled, however, there is still no indication when looking at the buttons that anything has changed.
My solution to this is that you should have the buttons have a different background color when enabled.
- Question & Suggestion for maintaining:
Have you thought about making this project open-source and turning it into a GitHub repository?
If your goal for this project is just to create a nice tool for the community, then I think sticking it to Github would be a great idea. That way you can have other members of the community help you with improving the project.
I hope these suggestions help you with your project.
Good Luck!
Thank you for your detailed comments.
I reworked the workflow and UI so now:
- LeaderBoard is updated every 2 hours
- Columns are marked hidden as you requested.
There are a bunch of new things I've added recently in the UI.
Hi, nice update!
Closing the issue since it's inactive atm and the resource is linked in the discussions resource :)
(we try to mostly keep actionable issues open)