7 ms·
Nice introduction, but I think that ranking the models purely by their input token limits is not a useful exercise. Looking at the MTEB leaderboard is better (a
by tomthe 2y ago
Nice introduction, but I think that ranking the models purely by their input token limits is not a useful exercise. Looking at the MTEB leaderboard is better (although a lot of the models are probably overfitting to their test set).
This is a good time to chill for my visualization of 5 Millionembeddings of HN posts, users and comments: https://tomthe.github.io/hackmap/ https://tomthe.github.io/hackmap/
- kaycebasques 2y agoThanks, a couple other people gave me this same feedback in another comment thread and it definitely makes sense not to overindex on input token size. Will update that section in a bit.