NVIDIA / TransformerEngine

A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit floating point (FP8) precision on Hopper and Ada GPUs, to provide better performance with lower memory utilization in both training and inference.
https://docs.nvidia.com/deeplearning/transformer-engine/user-guide/index.html
Apache License 2.0
1.85k stars 310 forks source link

Add the option to use SM for P2P comm in TP overlap #914

Closed erhoo82 closed 3 months ago

erhoo82 commented 3 months ago

Description

Add an option for SM-based P2P send/recv in TP communication overlap.

Type of change

Changes

Please list the changes introduced in this PR:

Checklist:

erhoo82 commented 3 months ago

Thanks @timmoon10. Updated the suggested changes.

timmoon10 commented 3 months ago

/te-ci pytorch

erhoo82 commented 3 months ago

@timmoon10 I think the paddle build failure is unrelated issue?

timmoon10 commented 3 months ago

/te-ci pytorch