VERIFYBrowser & web
Jev vs. OSS Model Evaluation

This project evaluates Jev and its OSS variants (Nimble, Bonsai) for design and data verification use cases, finding Jev and its variants to be less proficient in arithmetic operations compared to OSS models.
I evaluated Jev vs. OSS variants for my use cases
- design verification - does this site follow my design system?
- data verification - does our txn data match the chain public RPC data? (v important if u use it to trade lol)
tldr; shifting design verification to OSS Nimble 9B & data verification to Bonsai 2 27B (b/c Jev + its variants like Nimble bad at arithmetic it seems)
Jev might be better natively (i used free thru @vercel gateway). OSS models might also just be faster for me as I run them natively on my own hardware (& thus far free, even if Jev is supa cheap)
Check out the full breakdown here: yuanfenxyz.com/classification
I'm also testing out these models for X bookmark classification, which I run 2x a day on my GPU for my site! Will update that next

