Revert "Move torch pin from the 2.11 to the 2026-04-09 nightly, and drop deprecated CUDA versions from CI" by mergennachin · Pull Request #19160 · pytorch/executorch

mergennachin · 2026-04-27T18:11:24Z

Reverts #19072

Too many failures on

https://hud.pytorch.org/hud/pytorch/executorch/main/1?per_page=50

Lots of AOTI/CUDA/Metal failures

…rop depr…" This reverts commit d7f8718.

pytorch-bot · 2026-04-27T18:11:29Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19160

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

Rolling out OSDC (ARC) runners on pull & trunk workflows in PyTorch main

This comment was automatically generated by Dr. CI and updates every 15 minutes.

github-actions · 2026-04-27T18:12:10Z

This PR needs a `release notes:` label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.

Copilot

Pull request overview

This PR reverts ExecuTorch’s PyTorch dependency back toward the 2.11 release line (away from the 2026-04-09 nightly pin) to reduce CI instability, while updating related CI/config scripts and vendored c10 headers accordingly.

Changes:

Re-pin PyTorch to 2.11-era versions and switch several install paths from nightly to the PyTorch “test” wheel index.
Adjust vendored header-only c10 macros/guards (CUDA/ROCm warp size + BFloat16 CUDA guards) to match the reverted pin.
Update CI workflows and Docker/Windows CUDA cross-compile settings (CUDA matrix expansion; Windows CUDA baseline moved to 12.8).

Reviewed changes

Copilot reviewed 15 out of 15 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
`torch_pin.py`	Reverts TORCH_VERSION to 2.11.0 and comments out the nightly pin.
`runtime/core/portable_type/c10/torch/headeronly/util/BFloat16.h`	Simplifies CUDA-vs-ROCm guards around `__nv_bfloat16` usage.
`runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h`	Updates ROCm warp-size macro implementation and compiler-version gating logic.
`install_requirements.py`	Switches to PyTorch “test” index and hard-pins torch/vision/audio versions when “nightly” is enabled.
`examples/models/moshi/mimi/install_requirements.sh`	Pins torchcodec to 0.11.0 and uses the PyTorch “test” index.
`.github/workflows/docker-builds.yml`	Changes docker build matrix (notably runner sizing behavior for some images).
`.github/workflows/cuda.yml`	Expands CUDA build matrix to include 12.8 and 12.9.
`.github/workflows/cuda-windows.yml`	Moves CUDA Windows workflow baseline from 12.6 to 12.8.
`.ci/scripts/utils.sh`	Pins torchaudio/torchvision installs to release branches.
`.ci/scripts/test_wheel_package_qnn.sh`	Changes torch install to use a “test” index and removes nightly version concatenation.
`.ci/scripts/test_model_e2e.sh`	Pins torchcodec to 0.11.0 and uses the PyTorch “test” index.
`.ci/docker/common/install_pytorch.sh`	Updates torchaudio/torchvision refs to release branches.
`.ci/docker/common/install_cuda_windows_cross_compile.sh`	Updates supported CUDA versions and installer naming for Windows toolkit downloads.
`.ci/docker/ci_commit_pins/pytorch.txt`	Changes PyTorch pin from a SHA to `release/2.11`.
`.ci/docker/build.sh`	Updates CUDA Windows cross-compile Docker build to CUDA 12.8.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-27T18:19:39Z

      matrix:
        runner: [linux.4xlarge]
        docker-image-name: [
+          executorch-ubuntu-22.04-gcc11,
          executorch-ubuntu-22.04-gcc9-nopytorch,


The default runner for this matrix is linux.4xlarge, and this list now includes executorch-ubuntu-22.04-gcc11. That image builds PyTorch from source during the Docker build (SKIP_PYTORCH is not set for gcc11), which is very memory-intensive and has previously required larger runners to avoid OOM. Consider restoring an include: override to run the gcc11 image on a larger runner (or skipping the PyTorch-from-source build in this image).

Copilot · 2026-04-27T18:19:39Z

          executorch-ubuntu-22.04-linter,
          executorch-ubuntu-22.04-arm-sdk,
+          executorch-ubuntu-22.04-zephyr-sdk,
          executorch-ubuntu-22.04-qnn-sdk,
          executorch-ubuntu-22.04-mediatek-sdk,


executorch-ubuntu-22.04-zephyr-sdk is also on the default linux.4xlarge runner now, but it builds PyTorch from source during the Docker build (SKIP_PYTORCH is unset for zephyr-sdk). This is likely to OOM on 4xlarge and fail the docker-build workflow. Consider restoring a per-image runner override (e.g. linux.12xlarge) for this image, or otherwise avoiding the from-source PyTorch build here.

Copilot · 2026-04-27T18:19:40Z

+  # Install torch based on the pinned PyTorch version, preferring the PyTorch test index
+  "$PIPBIN" install torch=="${TORCH_VERSION}" --extra-index-url "https://download.pytorch.org/whl/test"


--extra-index-url https://download.pytorch.org/whl/test does not “prefer” the test index: pip will still prefer the primary index (PyPI) when the same version exists there. Also, other scripts use the per-platform subindex (e.g. .../whl/test/cpu or .../whl/test/cu126), and .../whl/test may not be the correct/simple index for wheels. If the goal is to force the test index, use --index-url with the correct subindex (and optionally add PyPI as an extra index), or update the comment to match actual behavior.

Suggested change

# Install torch based on the pinned PyTorch version, preferring the PyTorch test index

"$PIPBIN" install torch=="${TORCH_VERSION}" --extra-index-url "https://download.pytorch.org/whl/test"

# Install torch from the PyTorch test CPU wheel index, while allowing PyPI for dependencies

"$PIPBIN" install torch=="${TORCH_VERSION}" --index-url "https://download.pytorch.org/whl/test/cpu" --extra-index-url "https://pypi.org/simple"

Copilot · 2026-04-27T18:19:40Z

@@ -1 +1 @@
-358117c166b75167a09bca81ac9925940feda339
+release/2.11


Pinning to a moving ref (release/2.11) makes builds less reproducible and also conflicts with the existing pin-bump automation (.github/scripts/update_pytorch_pin.py) which writes a commit SHA into this file. Consider pinning to a specific commit hash (or at least a tag) and updating/pausing the automation if the intent is to stay on a stable release branch long-term.

Suggested change

release/2.11

v2.11.0

Copilot · 2026-04-27T18:19:40Z

        # Setting use_pytorch_nightly to false to test the pinned PyTorch commit. Note
        # that we don't need to set any version number there because they have already
        # been installed on CI before this step, so pip won't reinstall them
-        (
-            f"torch=={TORCH_VERSION}.{NIGHTLY_VERSION}"
-            if use_pytorch_nightly
-            else "torch"
-        ),
+        ("torch==2.11.0" if use_pytorch_nightly else "torch"),
    ]


This hard-codes the torch version string in two places (torch_pin.py and here). That duplication can drift over time and create inconsistent local vs CI installs. Consider importing/reading TORCH_VERSION from torch_pin.py and using it here instead of repeating 2.11.0.

Copilot · 2026-04-27T18:19:41Z

    print("Installing torch domain libraries")
    DOMAIN_LIBRARIES = [
-        (
-            f"torchvision==0.27.0.{NIGHTLY_VERSION}"
-            if use_pytorch_nightly
-            else "torchvision"
-        ),
-        (
-            f"torchaudio==2.11.0.{NIGHTLY_VERSION}"
-            if use_pytorch_nightly
-            else "torchaudio"
-        ),
+        ("torchvision==0.26.0" if use_pytorch_nightly else "torchvision"),
+        ("torchaudio==2.11.0" if use_pytorch_nightly else "torchaudio"),
    ]


These domain library versions are hard-coded here, which can drift from whatever torch/CI pin is intended and makes upgrades easy to miss. Consider centralizing the torchvision/torchaudio pins alongside the torch pin (or reusing the existing pin mechanism) so version bumps are consistent across scripts.

Revert "Move torch pin from the 2.11 to the 2026-04-09 nightly, and d…

78f6276

…rop depr…" This reverts commit d7f8718.

Copilot AI review requested due to automatic review settings April 27, 2026 18:11

mergennachin requested review from JacobSzwejbka and lucylq as code owners April 27, 2026 18:11

pytorch-bot Bot added the ci-no-td label Apr 27, 2026

meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Apr 27, 2026

mergennachin requested review from manuelcandales and rascani April 27, 2026 18:11

Copilot started reviewing on behalf of mergennachin April 27, 2026 18:12 View session

mergennachin requested a review from Gasoonjia April 27, 2026 18:12

rascani approved these changes Apr 27, 2026

View reviewed changes

Copilot AI reviewed Apr 27, 2026

View reviewed changes

mergennachin merged commit 3a62fac into main Apr 27, 2026
359 of 372 checks passed

mergennachin deleted the revert-19072-update_torch_nightly branch April 27, 2026 18:20

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Revert "Move torch pin from the 2.11 to the 2026-04-09 nightly, and drop deprecated CUDA versions from CI"#19160

Revert "Move torch pin from the 2.11 to the 2026-04-09 nightly, and drop deprecated CUDA versions from CI"#19160
mergennachin merged 1 commit intomainfrom
revert-19072-update_torch_nightly

mergennachin commented Apr 27, 2026

Uh oh!

pytorch-bot Bot commented Apr 27, 2026

Uh oh!

github-actions Bot commented Apr 27, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Copilot AI Apr 27, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		# Install torch based on the pinned PyTorch version, preferring the PyTorch test index
		"$PIPBIN" install torch=="${TORCH_VERSION}" --extra-index-url "https://download.pytorch.org/whl/test"

		@@ -1 +1 @@
		358117c166b75167a09bca81ac9925940feda339
		release/2.11 No newline at end of file

Conversation

mergennachin commented Apr 27, 2026

Uh oh!

pytorch-bot Bot commented Apr 27, 2026

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19160

❗ 1 Active SEVs

Uh oh!

github-actions Bot commented Apr 27, 2026

This PR needs a release notes: label

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 27, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

This PR needs a `release notes:` label