rpms/python-torch
3
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Better disabling of aotriton

Browse source 
Signed-off-by: Tom Rix <trix@redhat.com>
This commit is contained in:
Tom Rix 2024-03-19 14:53:39 -04:00
commit 74a2fd1d89
2 changed files with 47 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
python-torch.spec
View file

@ -98,6 +98,7 @@ Patch101: 0001-cuda-hip-signatures.patch
Patch102: 0001-silence-an-assert.patch
Patch103: 0001-can-not-use-with-c-files.patch
Patch104: 0001-use-any-hip.patch
Patch105: 0001-disable-use-of-aotriton.patch
%endif
ExclusiveArch: x86_64 aarch64
@ -278,8 +279,6 @@ sed -i -e '/fsspec/d' setup.py
# A new dependency
# Connected to USE_FLASH_ATTENTION, since this is off, do not need it
sed -i -e '/aotriton.cmake/d' cmake/Dependencies.cmake
# sed -i -e '/aotriton/d' aten/src/ATen/native/transformers/cuda/sdp_utils.cpp
rm aten/src/ATen/native/transformers/cuda/sdp_utils.cpp
# Release comes fully loaded with third party src
# Remove what we can