Skip to content

[torch] Adjust env vars used for builds with aotriton enabled. #1432

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ScottTodd merged 1 commit into from
Sep 9, 2025
Merged

[torch] Adjust env vars used for builds with aotriton enabled. #1432

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 2 additions & 9 deletions external-builds/pytorch/build_prod_wheels.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -362,6 +362,7 @@ def do_build(args: argparse.Namespace):
)

env: dict[str, str] = {
"PYTHONUTF8": "1", # Some build files use utf8 characters, force IO encoding
"CMAKE_PREFIX_PATH": str(cmake_prefix),
"ROCM_HOME": str(rocm_dir),
"ROCM_PATH": str(rocm_dir),
Expand Down Expand Up @@ -584,7 +585,7 @@ def do_build_pytorch(
# Add the _rocm_init.py file.
(pytorch_dir / "torch" / "_rocm_init.py").write_text(get_rocm_init_contents(args))

# Workaround missing features on windows.
# Windows-specific options.
if is_windows:
use_flash_attention = (
"1" if args.enable_pytorch_flash_attention_windows else "0"
Expand All @@ -593,14 +594,6 @@ def do_build_pytorch(
{
"USE_FLASH_ATTENTION": use_flash_attention,
"USE_MEM_EFF_ATTENTION": use_flash_attention,
# Currently, aotriton packages don't include windows binaries
# so we build them alongside pytorch using AOTRITON_INSTALL_FROM_SOURCE=1.
# On Windows, aotriton is built with "NOIMAGE" mode, so it needs kernel images built from Linux.
# TODO: TheRock provides aotriton artifacts compiled for windows including aotriton images built from Linux.
# For now, manually copy in the aotriton.images folder from linux binaries into <pytorch_root>/lib/aotriton.images.
# NOTE: this will not work without the corresponding patch in the main branch.
# which is in ./patches/pytorch/main/pytorch/hipified/0004-Support-FLASH_ATTENTION-MEM_EFF_ATTENTION-via.-aotriton.patch
"AOTRITON_INSTALL_FROM_SOURCE": use_flash_attention,
"DISTUTILS_USE_SDK": "1",
# Workaround compile errors in 'aten/src/ATen/test/hip/hip_vectorized_test.hip'
# on Torch 2.7.0: https://gist.github.com/ScottTodd/befdaf6c02a8af561f5ac1a2bc9c7a76.
Expand Down
Loading