16-bit quantization

#4
by semevas - opened

16-bit quantization?

The model uses bf16 by default. If you have an older GPU (like myself) and can't run bf16, try altering the dtype to "float16" in the model's config.json.

Sign up or log in to comment