LocalLLaMA

3 readers

1 users here now

Community to discuss about Llama, the family of large language models created by Meta AI.

founded 1 year ago

MODERATORS

communick@poweruser.forum

codellama gives really bad results, am I doing something wrong? (alien.top)

submitted 1 year ago by Detz@alien.top to c/localllama@poweruser.forum

4 comments fedilink hide all child comments

Is this modal just not good enough? I would think it would know the difference between python and js at the very least?

https://preview.redd.it/xf06g5dn8b0c1.png?width=1754&format=png&auto=webp&s=fd4d297400f379bd7d679d963359f550242a8de4

you are viewing a single comment's thread
view the rest of the comments

[–] SomeOddCodeGuy@alien.top 1 points 1 year ago

I'll be honest- my expectation of a 7b q6 model to generate a usable code block would be pretty darn low. I use 7b as a small code snippet analyzer in Continue.dev, but I'm not sure I'd really put a lot of faith in any 7b to do that task. The CodeLlama 34b models are actually pretty decent, though; I use the 34b q8 of CodeFuse, Codebooga and Phind v2 and they are all pretty decent.