microsoft / microxcaling

PyTorch emulation library for Microscaling (MX)-compatible data formats
MIT License
123 stars 14 forks source link

Could You Offer a More Comprehensive Environment Setup Guide? #16

Open Mr-Philo opened 5 months ago

Mr-Philo commented 5 months ago

Hi, firstly I want to extend my sincere thanks for the work you've done on MXFP. It has been a valuable resource for the community.

I am reaching out to request an update to the installation instructions provided in the README file. Currently the README outlines the basic requirements for the setup, particularly the necessity of CUDA (11.3+ recommended) and a reference to the requirements.txt file for Python packages.

However, README doesn't provide a complete installation procedure, especially for the cpp extension. Add the pytorch version, CUDA version, g++ version may confict with each other if not specified. There may be problems when users trying to build custom CUDA extensions using python build.py install.

Besides, it would be nice to have mirrors or docker containers with complete environments.

I believe that a more detailed installation guide would be greatly beneficial, especially for those who want to look deeper into MXFP. Thank you again!