why fp8_e4m3 min_scaling_factor divide 512? #3590

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Open

HPC4AI opened this issue Jan 20, 2025 · 0 comments

HPC4AI commented Jan 20, 2025

Hello, I would like to perform quantization from the FP16 data type to the FP8E4M3 data type. I referred to the method described in the link https://github.com/pytorch/FBGEMM/blob/main/fbgemm_gpu/experimental/gen_ai/src/quantize/quantize.cu#L629, but I have a question. Why is the calculation of min_scaling_factor done by dividing by (FP8_E4M3_MAX::value * 512.f)? Could you please explain the basis for choosing 512.f? Thanks.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment