logoalt Hacker News

trentoryesterday at 11:32 PM0 repliesview on HN

Yes it is slow for training/fine-tuning decently sized LLMs and unusable for diffusion models. What are you training were you get usable speeds?