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 309 forks source link

Change condition for ub tp overlap. #1055

Open Victarry opened 2 months ago

Victarry commented 2 months ago

Description

For MoELayer in MCore, we set parallel_mode=None to explicitly handle communications between TPxEP group. When using --tp-comm-overlap, the ub_overlap_rs will be enabled even when parallel_mode=None, which caused wrong results.

Fixes # (issue)

Type of change

Changes

Please list the changes introduced in this PR:

Checklist: