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

[C/PyTorch/JAX] Build system improvements for rpath and C++11 ABI #858

Closed denera closed 3 months ago

denera commented 4 months ago

Description

This PR improves how rpaths are handled for framework extensions, and makes the core TE library inherit the C++ ABI version setting from PyTorch when TE is being built for PyTorch integration.

Fixes # (issue)

Type of change

Changes

Checklist:

denera commented 4 months ago

/te-ci Pytorch

denera commented 4 months ago

/te-ci jax

denera commented 4 months ago

/te-ci pytorch

denera commented 4 months ago

/te-ci jax

denera commented 4 months ago

/te-ci paddle

denera commented 4 months ago

/te-ci paddle

denera commented 4 months ago

@ptrendx @ksivaman @timmoon10 CI is now clean for all three frameworks. I held off from merging because it's a significant build system change. Please let me know if it's clear to merge, or go ahead and merge whenever it is appropriate.

denera commented 4 months ago

@ksivaman No objections on my end. I think it makes sense to combine.

ksivaman commented 3 months ago

@denera Closing in favor of https://github.com/NVIDIA/TransformerEngine/pull/877

timmoon10 commented 3 months ago

:(