Repository navigation
torchao release compatibility table #2919
Description
Activity
any ideas when we can fix the compatibility of torchao nightly with torch? currently it's blocking tests in vllm: https://github.com/vllm-project/vllm/pull/21982/files#diff-2fe466060a88bb6a57175df8ca7175849db82a2cf2ba082295d481ab57e58868R512
I'm also having the same issue here. Blocking testing executorch with torch nightly.
I have torch on latest master (with some python-only local changes)
'2.9.0a0+git3564a8a', and torchaotorchao-0.14.0.dev20250909+cu126.Reacted by Jerry ZhangI'm also having the same issue here. Blocking testing executorch with torch nightly.
I have torch on latest master (with some python-only local changes)
'2.9.0a0+git3564a8a', and torchaotorchao-0.14.0.dev20250909+cu126.I have it resolved by installing
pip install fbgemm-gpu-nightlyto overridefbgemm-gpu. So far importing torchao doesn't error anymore. Not sure if there're any other issues if I actually run anything.executorch doesn't need fbgem-gpu I think? if you use torchao nightly, can you also use torch nightly in ET?
Reacted by Austinwe also need to update the following check https://github.com/pytorch/ao/blob/ea8c00fc90c99f0bf19fe87d22eb186c3dd19bf6/torchao/__init__.py#L38C45-L38C76 for PyTorch 2.10.x, as
str(torch.__version__) >= "2.9"will not work properly for PyTorch 2.10@liangel-02 , https://github.com/facebookresearch/FBGEMM/pull/1900/files might be relevant here - this is fbgemm fixing the same issue in their repo
Getting the error on all versions:
Skipping import of cpp extensions due to incompatible torch version 2.9.0+cu128 for torchao version 0.14.0
Skipping import of cpp extensions due to incompatible torch version 2.8.0+cu129 for torchao version 0.14.0
Skipping import of cpp extensions due to incompatible torch version 2.8.0+cu128 for torchao version 0.14.0Reacted by Mauro Sciancalepore, Kirill Buev, namgyu-youn, Tianang Leng, Thijs, Tensor Templar, Siyuan Liu, Nguyen Anh Mai, juejuezi7, Le Nho Bach and 5 moreThis makes it very hard to use torchao with latest versions of Ray Serve/vLLM:
Skipping import of cpp extensions due to incompatible torch version 2.8.0+cu128 for torchao version 0.14.0Reacted by Patrick von PlatenHi folks, thank you for reporting, we are looking into this and will provide an update soon. Note that unless you actually need c++ or CUDA kernels that ship with torchao, you can ignore the warning and use the Python-only APIs without issues.
Reacted by Kirill BuevPyTorch release 2.9.0 eta date is today : https://dev-discuss.pytorch.org/t/pytorch-2-9-final-rc-available/3245
still have this warning message for torch 2.9.0:
Skipping import of cpp extensions due to incompatible torch version 2.9.0+cu128 for torchao version 0.14.0 Please see GitHub issue #2919 for more info
These frustrating warnings keeps appearing repeatedly on all envs I working. Please fix it or mute it
Reacted by Marco De Nadai and Himanshu mauryaHi all, we plan to release a 0.14.1 as soon as possible. The new version will be built against 2.9.0 and it will load the cpp extensions when used against this version. You won't see this warning anymore.
Reacted by Steve, Jiameng Gao, namgyu-youn, Matthijs Van keirsbilck and DefTruthAre there any plans to support CUDA 13.0? Thank you.
44 remaining items
- added a commit that references this issue
on Mar 24, 2026 - added a commit that references this issue
on May 4, 2026 @vkuzo please update:
# torchao == 0.16.0 torchao_pytorch_compatible_versions = [ # Built against torch 2.8.0 (_parse_version("0.13.0"), _parse_version("2.8.0")), (_parse_version("0.14.0"), _parse_version("2.8.0")), # Built against torch 2.9.0 (_parse_version("0.14.1"), _parse_version("2.9.0")), (_parse_version("0.14.1"), _parse_version("2.10.0.dev")), # Built against torch 2.9.1 (_parse_version("0.15.0"), _parse_version("2.9.1")), (_parse_version("0.15.0"), _parse_version("2.10.0.dev")), # Built against torch 2.10.0 (_parse_version("0.16.0"), _parse_version("2.10.0")), (_parse_version("0.16.0"), _parse_version("2.11.0.dev")), # current torchao version - to be deleted after version bump (_parse_version("0.16.0.dev"), _parse_version("2.9.1")), (_parse_version("0.16.0.dev"), _parse_version("2.10.0.dev")), (_parse_version("0.16.0.dev"), _parse_version("2.11.0.dev")), # next torchao version (_parse_version("0.17.0.dev"), _parse_version("2.10.0")), (_parse_version("0.17.0.dev"), _parse_version("2.11.0.dev")), ]The latest version to support torch 2.8.0cu is "0.13.0" (training platform in Hopper with stable support of cu12.8).
Closing this issue since torchao 0.17.0 is now ABI stable against torch 2.11. We have removed the compatibility table for all future torchao versions starting with 0.17.0. Please refer to https://github.com/pytorch/ao/releases/tag/v0.17.0 for more details.
- added a commit that references this issue
on May 13, 2026 @andrewor14 for Torch 2.12.0 we can use 0.17.0? what should we use? there is stable torch 2.12.0
This issue describes the compatibility matrix between torchao releases and its dependencies. If you are seeing an error when importing
torchaothat looks like this,(pytorch_nightly) [vasiliy@devgpu007.eag6 ~/local]$ python -c "import torchao"
Fatal Python error: Aborted
then most likely you can resolve this error by ensuring that thetorchversion in your environment is compatible with thetorchversion used to build yourtorchaoversion.torch
torchao version torch version torch version, torchao's Python API only
0.18.0dev (nightly) 2.12.0dev (nightly) 2.11.0, 2.10.0, 2.9.0
0.17.0 2.11.0+ 2.11.0, 2.10.0, 2.9.0
0.16.0 2.10.0 2.10.0, 2.9.0, 2.8.0
0.15.0 2.9.1 2.9.1, 2.8.0, 2.7.1
0.14.1 2.9.0 2.9.0, 2.8.0, 2.7.1
0.13.0 2.8.0 2.8.0, 2.7.1, 2.6.0
0.12.0 2.7.1, 2.6.0, 2.5.0 n/afbgemm_gpu
torchaohas an optional runtime dependency onfbgemm_gpu. Please see https://docs.pytorch.org/FBGEMM/general/Releases.html for the compatibility matrix forfbgemm_gpu. Note that whiletorchao's Python API supports multipletorchversions, eachfbgemm_gpuversion only supports a singletorchversion. Therefore, if you are usingtorchaotogether withfbgemm_gpu, you should use thetorchversion corresponding to yourfbgemm_gpuversion.you are best
- added a commit that references this issue
on Jun 17, 2026 you really should start releasing cuda 13 windows wheels
with CODEX I was able to compile abi3 torch 2.13 cuda for windows for my followers
- unpinned this issue
on Aug 10, 2026 - added a commit that references this issue
on Aug 29, 2026 - added 2 commits that reference this issue
on Sep 8, 2026
This issue describes the compatibility matrix between torchao releases and its dependencies. If you are seeing an error when importing
torchaothat looks like this,then most likely you can resolve this error by ensuring that the
torchversion in your environment is compatible with thetorchversion used to build yourtorchaoversion.torch
fbgemm_gpu
torchaohas an optional runtime dependency onfbgemm_gpu. Please see https://docs.pytorch.org/FBGEMM/general/Releases.html for the compatibility matrix forfbgemm_gpu. Note that whiletorchao's Python API supports multipletorchversions, eachfbgemm_gpuversion only supports a singletorchversion. Therefore, if you are usingtorchaotogether withfbgemm_gpu, you should use thetorchversion corresponding to yourfbgemm_gpuversion.