Hacker News new | ask | show | jobs
by pzo 461 days ago
Maybe model is sensitive to quantization, by default ollama quantize it significantly.
1 comments
tarruda 461 days ago
I tried ollama fp16 and it had the same issues.
link