Commits on Sep 30, 2024

1. Make some fbgemm fp8 triton ops pt2 friendly (pytorch#3188) …

   Summary:
   X-link: facebookresearch/FBGEMM#283
   
   Pull Request resolved: pytorch#3188
   
   Make some fbgemm fp8 triton ops pt2 friendly..
   
   # What this diff tries to do
   * stop using TensorWrapper and tl.reinterpret
   * Remove the use of triton_heuristics for _kernel_matmul_fp8_row
   
   # What this diff won't help:
   * triton_herustics use cases of EVEN_K. One option is to just merge that into the autotuning configs
   
   # need to do in the future:
   * Update other ops, like quantize_fp8_row.
   * Update documentation. Feels pretty outdated, and some still reference to TensorWrapper.
   
   Reviewed By: henrylhtsang
   
   Differential Revision: D63560103

   

   jwfromm authored and facebook-github-bot committed Sep 30, 2024
   Configuration menu

   • View commit details
   • Copy full SHA for be457d5
   • Browse repository at this point

   Copy the full SHA

   be457d5 View commit details

   Browse the repository at this point in the history