Nastya
@nastya
Okay, the benchmark results may not be that good https://www.reddit.com/r/LocalLLaMA/comments/1g5srfa/no_the_llama31nemotron70binstruct_has_not_beaten/
3 replies
0 recast
2 reactions
nemb
@nemb
Depends on what you do. o1 is honestly awful for quite a few things. Nemotron pretty good at a few things (especially if you CoT-prompt it like o1 does). Sonnet 3.5 still king for code. Qwen2.5-coder, Deepseek-Coder V2 16B, Codegeex in the open source category if you don't need a too large scope.
0 reply
0 recast
1 reaction
koisose.lol
@koisose
still need to use o1-preview then
0 reply
0 recast
1 reaction
Rafi
@rafi
Who'd've thought that folks were optimistic about performance of the model before they actually rolled it out 🤔
0 reply
0 recast
1 reaction
