Replying to kpr797

I've been playing around with running #deepseek R1 (the 7 billion parameters model) locally on my computer.

While it's cool that it can be run locally, the answers it provides are not particularly great.

How many billion parameters does it take to start getting consistently good answers?

#asknostr #ai #llm
Avatar
Logen 11mo ago

I don’t get great answers either. I find that it hallucinates and contracts itself quite often. I’ve had much better results with phi4 or llama 3.1

Reply to this note

Please Login to reply.

Discussion

ec
kpr797 11mo ago

Thanks. I'll check those ones out.