8 comments

[ 3.8 ms ] story [ 35.8 ms ] thread
Yes! I’ve been waiting for this
Hey everyone! As part of running ModelDepot, I get a ton of requests for models that might solve a specific problem. Unfortunately the current volume of curated ML models on ModelDepot wasn't nearly enough so I thought if there was a better way to discover the ones that already exist out there, that could go a long way and help people use ML even easier (and hopefully lower the amount of 2am chats that are asking for a specific model haha).

I've spent the last month learning how to build a search engine from nothing, and had a lot of fun creating various ML pipelines and tuning Elasticsearch to surface good results on everything :)

I'd love to hear any feedback you guys have on different search patterns it's missing, genre of models it hasn't indexed yet, weird result rankings, or anything else in general! I'm also always free to chat via the in-site chat as well :)

what are the criteria for aggregating your models?
It starts off with a set of searches on Github like Tensorflow, ICCV, etc. and then from there it classifies repos as ML related or not, and then continues to crawl referenced repos from the ones that are ML related. Would love to hear if you know of any other interesting sources :)
This is excellent. Just signed up for community.