Update torch pin to 2.11 release candidate by mergennachin · Pull Request #18287 · pytorch/executorch

mergennachin · 2026-03-18T11:41:16Z

Trying to pin to nightly 3.18 but the some linux cuda wheels weren't successfully built last night.

Need to pin to stable release candidate due to time criticality.

Will need to revert this PR and pin normally to torch nightlies

Fix parakeet/voxtral realtime export using PyTorch release. In both `export_parakeet_tdt.py` and `export_voxtral_rt.py`, replaced the manual decomposition dictionary with `torch.export.default_decompositions()` and added the custom `_linear_bias_decomposition` for `torch.ops.aten.linear.default`.

pytorch-bot · 2026-03-18T11:41:21Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18287

📄 Preview Python docs built from this PR

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

❌ 1 Awaiting Approval, 1 New Failure, 9 Cancelled Jobs, 3 Unrelated Failures

As of commit af00588 with merge base b1373e8 ():

AWAITING APPROVAL - The following workflow needs approval before CI can run:

Claude Code (gh)

NEW FAILURE - The following job has failed:

pull / test-multimodal-linux (gemma3-4b) / linux-job (gh)
RuntimeError: Command docker exec -t a6c7bd2f43a2842cc4411d24ecebf6479072e297b9180176d4fae9dd005f92d6 /exec failed with exit code 139

CANCELLED JOBS - The following jobs were cancelled. Please retry:

pull / test-lora-linux / linux-job (gh)
##[error]The operation was canceled.
pull / test-samsung-models-linux / linux-job (gh)
##[error]The operation was canceled.
pull / unittest / linux / linux-job (gh)
##[error]The operation was canceled.
pull / unittest-editable / linux / linux-job (gh)
##[error]The operation was canceled.
pull / unittest-nxp-neutron / linux-job (gh)
##[error]The operation was canceled.
trunk / test-demo-backend-delegation (buck2) / linux-job (gh)
##[error]The operation was canceled.
trunk / test-demo-backend-delegation (cmake) / linux-job (gh)
##[error]The operation was canceled.
trunk / test-models-linux-aarch64 (ic4, portable, linux.arm64.2xlarge) / linux-job (gh)
##[error]The operation was canceled.
trunk / unittest-release / linux / linux-job (gh)
##[error]The operation was canceled.

BROKEN TRUNK - The following jobs failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / unittest / windows / windows-job (gh) (trunk failure)
##[error]The operation was canceled.
pull / unittest-editable / windows / windows-job (gh) (trunk failure)
##[error]The operation was canceled.
trunk / unittest-release / windows / windows-job (gh) (trunk failure)
##[error]The operation was canceled.

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

github-actions · 2026-03-18T11:42:00Z

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

Updates the repository’s PyTorch integration toward the 2.11 release/RC line, adjusts dependency installation sources, and incorporates a ROCm-specific complex math workaround alongside export decomposition fixes for Metal-related flows.

Changes:

Switch PyTorch install flows to use the PyTorch “test” wheel index and pin torch/vision/audio versions for that path.
Update model export scripts (parakeet/voxtral realtime) to use torch.export.default_decompositions() plus a custom linear-bias decomposition.
Add a HIP-specific complex<float> pow implementation intended to mitigate a ROCm 7.2 numerical regression.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
torch_pin.py	Adjusts torch pin metadata (nightly pin no longer a live Python constant).
runtime/core/portable_type/c10/c10/util/complex_math.h	Adds HIP-specific `pow` specialization for `complex<float>` using FMA-based multiplication.
install_requirements.py	Moves install base URL to PyTorch “test” index and hardcodes torch/vision/audio versions for the “nightly” path.
exir/sym_util.py	Uses `guarding_hint_or_throw` when available, falling back to deprecated `size_hint`.
examples/models/voxtral_realtime/export_voxtral_rt.py	Uses default decompositions table with a custom linear decomposition override.
examples/models/parakeet/export_parakeet_tdt.py	Uses default decompositions table with a custom linear decomposition override.
.ci/scripts/utils.sh	Pins torchaudio/torchvision to release branches instead of commit-pin files.
.ci/scripts/test_wheel_package_qnn.sh	Switches torch install to PyTorch “test” index and removes NIGHTLY_VERSION usage.
.ci/docker/common/install_pytorch.sh	Pins torchaudio/torchvision to release branches in Docker build.
.ci/docker/ci_commit_pins/pytorch.txt	Changes PyTorch pin from a commit SHA to `release/2.11`.

Comments suppressed due to low confidence (1)

install_requirements.py:28

install_requirements.py still describes "nightly" behavior and references NIGHTLY_VERSION/hud.pytorch.org nightly pins, but the script now installs from the /whl/test index and no longer uses NIGHTLY_VERSION. Please update these top-of-file notes to match the new release/RC/test-index workflow so users don’t follow stale instructions.

# The pip repository that hosts nightly torch packages.
# This will be dynamically set based on CUDA availability and CUDA backend enabled/disabled.
TORCH_URL_BASE = "https://download.pytorch.org/whl/test"

# Since ExecuTorch often uses main-branch features of pytorch, only the nightly
# pip versions will have the required features.
#
# NOTE: If a newly-fetched version of the executorch repo changes the value of
# NIGHTLY_VERSION, you should re-run this script to install the necessary
# package versions.
#
# NOTE: If you're changing, make the corresponding change in .ci/docker/ci_commit_pins/pytorch.txt
# by picking the hash from the same date in
# https://hud.pytorch.org/hud/pytorch/pytorch/nightly/ @lint-ignore

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

You can also share your feedback on Copilot code review. Take the survey.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot

Pull request overview

This PR temporarily shifts ExecuTorch’s PyTorch dependency strategy from a date-stamped nightly pin to a PyTorch 2.11 release-candidate/stable track (motivated by missing/broken nightly CUDA wheels), and updates a ROCm-specific complex math implementation to avoid a ROCm 7.2 regression.

Changes:

Switch installation logic to use the PyTorch whl/test index and pin torch/torchvision/torchaudio to 2.11/0.26 versions.
Update CI/docker scripts to use release/* refs for PyTorch domain libraries and set the PyTorch pin file to release/2.11.
Add a HIP-specific pow(complex<float>, complex<float>) specialization using an FMA-preserving multiply to mitigate ROCm 7.2 numerical regression.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
`torch_pin.py`	Comments out `NIGHTLY_VERSION` to reflect temporary non-nightly pin.
`runtime/core/portable_type/c10/c10/util/complex_math.h`	Adds HIP specialization for complex pow to address ROCm regression.
`install_requirements.py`	Points installs at `whl/test` and hard-pins torch/domain versions.
`exir/sym_util.py`	Uses `guarding_hint_or_throw` when available, with fallback to deprecated `size_hint`.
`examples/models/voxtral_realtime/export_voxtral_rt.py`	Uses `default_decompositions()` table and overrides linear decomposition.
`examples/models/parakeet/export_parakeet_tdt.py`	Same decomposition-table approach for Metal export.
`.ci/scripts/utils.sh`	Pins audio/vision to `release/*` refs instead of commit pins.
`.ci/scripts/test_wheel_package_qnn.sh`	Stops composing nightly torch version; installs `torch==${TORCH_VERSION}` from `whl/test` (extra index).
`.ci/docker/common/install_pytorch.sh`	Pins audio/vision to `release/*` refs in Docker build.
`.ci/docker/ci_commit_pins/pytorch.txt`	Changes PyTorch pin from a SHA to `release/2.11`.

Comments suppressed due to low confidence (1)

install_requirements.py:69

torch_url is derived from the PyTorch wheel index (CPU/CUDA via determine_torch_url), but it’s passed via --extra-index-url. Now that torch==2.11.0 exists on PyPI, pip may prefer the PyPI (CPU) wheel over the CUDA wheel from download.pytorch.org, and it also won’t reliably select a release-candidate from the whl/test index. To ensure the intended wheel is installed, prefer making torch_url the primary index (--index-url) for the torch/vision/audio installs (and keep PyPI as an extra index for other deps).

    torch_url = determine_torch_url(TORCH_URL_BASE)

    # pip packages needed by exir.
    TORCH_PACKAGE = [
        # 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
        ("torch==2.11.0" if use_pytorch_nightly else "torch"),
    ]

    # Install the requirements for core ExecuTorch package.
    # `--extra-index-url` tells pip to look for package
    # versions on the provided URL if they aren't available on the default URL.
    subprocess.run(
        [
            sys.executable,
            "-m",
            "pip",
            "install",
            "-r",
            "requirements-dev.txt",
            *TORCH_PACKAGE,
            "--extra-index-url",
            torch_url,
        ],

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

You can also share your feedback on Copilot code review. Take the survey.

 # The pip repository that hosts nightly torch packages.
 # This will be dynamically set based on CUDA availability and CUDA backend enabled/disabled.
-TORCH_NIGHTLY_URL_BASE = "https://download.pytorch.org/whl/nightly"
+TORCH_URL_BASE = "https://download.pytorch.org/whl/test"

 # Since ExecuTorch often uses main-branch features of pytorch, only the nightly
 # pip versions will have the required features.


+  echo "=== [$LABEL] Install torch==${TORCH_VERSION} ==="
+
+  # 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"


@@ -1,2 +1,2 @@
 TORCH_VERSION = "2.11.0"
-NIGHTLY_VERSION = "dev20260215"
+# NIGHTLY_VERSION = "dev20260318" Temporarily pinning to stable release candidate. Revert https://github.com/pytorch/executorch/pull/18287


@@ -1,2 +1,2 @@
 TORCH_VERSION = "2.11.0"
-NIGHTLY_VERSION = "dev20260215"
+# NIGHTLY_VERSION = "dev20260318" Temporarily pinning to stable release candidate. Revert https://github.com/pytorch/executorch/pull/18287


@@ -1 +1 @@
-659af3c353e49b35c191cdd2dba3b3c79d0e6822
+release/2.11


+  TORCHAUDIO_VERSION=release/2.11
  export TORCHAUDIO_VERSION
-  TORCHVISION_VERSION=$(cat .github/ci_commit_pins/vision.txt)
+  TORCHVISION_VERSION=release/0.26
  export TORCHVISION_VERSION


+  TORCHAUDIO_VERSION=release/2.11
  export TORCHAUDIO_VERSION
-  TORCHVISION_VERSION=$(cat .github/ci_commit_pins/vision.txt)
+  TORCHVISION_VERSION=release/0.26
  export TORCHVISION_VERSION


@@ -114,20 +108,12 @@ def install_requirements(use_pytorch_nightly):

 def install_optional_example_requirements(use_pytorch_nightly):
    # Determine the appropriate PyTorch URL based on CUDA delegate status
-    torch_url = determine_torch_url(TORCH_NIGHTLY_URL_BASE)
+    torch_url = determine_torch_url(TORCH_URL_BASE)

    print("Installing torch domain libraries")
    DOMAIN_LIBRARIES = [
-        (
-            f"torchvision==0.26.0.{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"),
    ]


This reverts commit b40d6fe.

torch_pin.py is now the single source of truth for torch + the three domain libraries (vision/audio/codec). It exposes a CHANNEL field (nightly/test/release), the four version constants, NIGHTLY_VERSION, and helpers — torch_spec() / torchaudio_spec() / torchcodec_spec() / torchvision_spec() emit the right pip spec, torch_index_url_base() returns the right wheel index, and torch_branch() derives the upstream release/M.N branch from TORCH_VERSION. Every consumer — install_requirements.py, the two install_pytorch.sh / utils.sh shell helpers, test_model_e2e.sh, test_wheel_package_qnn.sh, the moshi/mimi install_requirements.sh, the update_pytorch_pin.py script, and the weekly bump workflow — reads through these helpers instead of re-encoding the version strings. For test/release the docker image build (install_pytorch.sh) and the CI bring-up path (utils.sh) install the published wheels directly via pip --index-url, replacing the source-build-from-git-branch dance inherited from the original 2.11 RC pin (#18287). Nightly keeps the source-build path so CI catches upstream regressions. This commit is a no-op for the active pin: CHANNEL = "test" with TORCH_VERSION = "2.11.0", TORCHAUDIO_VERSION = "2.11.0", TORCHCODEC_VERSION = "0.11.0", TORCHVISION_VERSION = "0.26.0" produces the same wheels (from /whl/test/cpu) that origin/main currently hardcodes. Switching channels later (e.g. to nightly post-release) becomes a one-line change in torch_pin.py plus bumping the version constants. update_pytorch_pin.py imports CHANNEL / NIGHTLY_VERSION / torch_branch directly (no more regex parsing of the file). For nightly it pins to an immutable SHA looked up by date; for test/release it writes torch_branch() (e.g. "release/2.11") into .ci/docker/ci_commit_pins/pytorch.txt so git checkout follows cherry-picks as they land. The weekly-pytorch-pin-bump workflow is guarded on CHANNEL == "nightly" and uses an in-place re.sub on NIGHTLY_VERSION (the previous \`printf '...' > torch_pin.py\` would have clobbered the new constants and helpers). test/test_torch_pin.py covers all three channels, all four specs, and the release/M.N branch derivation. Co-authored-by: Claude <noreply@anthropic.com>

mergennachin and others added 4 commits March 18, 2026 07:05

Update torch pin nightly to 20260310

f4c87c1

Update to 3.18

96b20df

Update torch pin to release

131f83f

mergennachin requested review from JacobSzwejbka, larryliu0820 and lucylq as code owners March 18, 2026 11:41

Copilot AI review requested due to automatic review settings March 18, 2026 11:41

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 Mar 18, 2026

Copilot started reviewing on behalf of mergennachin March 18, 2026 11:41 View session

mergennachin added ciflow/metal ciflow/cuda labels Mar 18, 2026

mergennachin requested a review from manuelcandales March 18, 2026 11:42

Add comment

fad6df9

Copilot AI reviewed Mar 18, 2026

View reviewed changes

Potential fix for pull request finding

af00588

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings March 18, 2026 11:53

Copilot started reviewing on behalf of mergennachin March 18, 2026 11:53 View session

Copilot AI reviewed Mar 18, 2026

View reviewed changes

manuelcandales approved these changes Mar 18, 2026

View reviewed changes

Use torchcodec 0.11.0 RC

b001bcd

mergennachin merged commit b40d6fe into main Mar 18, 2026
349 of 358 checks passed

mergennachin deleted the mergennachin/update_to_release_candidate branch March 18, 2026 13:48

mergennachin temporarily deployed to upload-benchmark-results March 18, 2026 14:45 — with GitHub Actions Inactive

mergennachin mentioned this pull request Apr 23, 2026

Move torch pin from the 2.11 to the 2026-04-09 nightly, and drop deprecated CUDA versions from CI #19072

Merged

mergennachin added a commit that referenced this pull request Apr 24, 2026

Revert "Update torch pin to 2.11 release candidate (#18287)"

ab00b01

This reverts commit b40d6fe.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update torch pin to 2.11 release candidate#18287

Update torch pin to 2.11 release candidate#18287
mergennachin merged 7 commits intomainfrom
mergennachin/update_to_release_candidate

mergennachin commented Mar 18, 2026 •

edited

Loading

Uh oh!

pytorch-bot Bot commented Mar 18, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Mar 18, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		@@ -1 +1 @@
		659af3c353e49b35c191cdd2dba3b3c79d0e6822
		release/2.11 No newline at end of file

Conversation

mergennachin commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot Bot commented Mar 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18287

❌ 1 Awaiting Approval, 1 New Failure, 9 Cancelled Jobs, 3 Unrelated Failures

Uh oh!

github-actions Bot commented Mar 18, 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!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mergennachin commented Mar 18, 2026 •

edited

Loading

pytorch-bot Bot commented Mar 18, 2026 •

edited

Loading

This PR needs a `release notes:` label