Add nightly C++11 ABI builds #8307
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
I'm not sure how to test this change. I've sent out the |
bhavya01
approved these changes
Oct 25, 2024
tengyifei
force-pushed
the
cxx11-builds
branch
from
7c5cc22
to
9a1f121
Compare
Initially, PyTorch/XLA was built with C++11 ABI, which was a more modern ABI (e.g. small string optimizations). However, that conflicts with the ABI of upstream PyTorch, so #5650 disabled the C++11 ABI. It turns out that C++11 ABI improves LazyTensor tracing performance drastically for some large models. We can't go back to C++11 ABI due to conflict with upstream, so this PR adds additional nightly build variants that enable C++11 ABI next to the existing wheels and dockers. The docker images will be tagged like "nightly_3.10_tpuvm_cxx11_20241023" instead of "nightly_3.10_tpuvm_20241023". The wheels will be named like "torch_xla-2.6.0.dev20241023.cxx11-cp310-cp310-linux_x86_64.whl" instead of "torch_xla-2.6.0.dev20241023-cp310-cp310-linux_x86_64.whl". This PR also adds support for building C++11 ABI variants for stable branches, but we don't activate it yet. This PR also removes Python 3.8 nightly builds. Fixes #8306.
tengyifei
force-pushed
the
cxx11-builds
branch
from
9a1f121
to
0228442
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Initially, PyTorch/XLA was built with C++11 ABI, which was a more modern ABI (e.g. small string optimizations). However, that conflicts with the ABI of upstream PyTorch, so #5650 disabled the C++11 ABI. It turns out that C++11 ABI improves LazyTensor tracing performance drastically for some large models. We can't go back to C++11 ABI due to conflict with upstream, so this PR adds additional nightly build variants that enable C++11 ABI next to the existing wheels and dockers.
The docker images will be tagged like "nightly_3.10_tpuvm_cxx11_20241023" instead of "nightly_3.10_tpuvm_20241023".
The wheels will be named like
"torch_xla-2.6.0.dev20241023.cxx11-cp310-cp310-linux_x86_64.whl" instead of "torch_xla-2.6.0.dev20241023-cp310-cp310-linux_x86_64.whl".
This PR also adds support for building C++11 ABI variants for stable branches, but we don't activate it yet.
Fixes #8306.