Repository navigation
[release/1.6] Fix remaining CI failures and release runner refresh - #24338
Merged
Merged
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24338
Note: Links to docs will display an error until the docs builds have been completed. ❌ 46 New Failures, 1 Cancelled Job, 1 Pending, 3 Unrelated Failures, 17 Unclassified FailuresAs of commit d3add7b with merge base 4db778e ( NEW FAILURES - The following jobs have failed:
UNCLASSIFIED FAILURES - DrCI could not classify the following jobs because the workflow did not run on the merge base. The failures may be pre-existing on trunk or introduced by this PR:
CANCELLED JOB - The following job was cancelled. Please retry:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
JacobSzwejbka
deployed
to
upload-benchmark-results
October 30, 2026 16:13 — with 
GitHub Actions
Active
GitHub Actions
Active
Compile the Triton kernels during AOTI lowering: with autotune_at_compile_time OFF, AOTInductor also generates JIT code and the JIT code fails to run on gfx950 with the pinned ROCm 7.14 toolchain, so lowering never completes. Verify the generated code object with the bare architecture name instead of the full amdgcn triple, which the toolchain does not embed in the program or the hsaco files, and require a compiled hsaco for the device. A diagnostic run confirmed the .pte and every hsaco carry bare gfx950 while the 256-byte .ptd carries neither.
JacobSzwejbka
had a problem deploying
to
cadence
October 30, 2026 18:17 — with
GitHub Actions
Failure
GitHub Actions
Failure
JacobSzwejbka
had a problem deploying
to
cadence
October 30, 2026 18:18 — with
GitHub Actions
Failure
GitHub Actions
Failure
JacobSzwejbka
had a problem deploying
to
cadence
October 30, 2026 18:19 — with
GitHub Actions
Failure
GitHub Actions
Failure
JacobSzwejbka
had a problem deploying
to
upload-benchmark-results
October 30, 2026 18:56 — with
GitHub Actions
Failure
GitHub Actions
Failure
shoumikhin
approved these changes
Oct 30, 2026
This file contains hidden or 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
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.
Fix the CI failures remaining after #24199 on release/1.6.
The installer test now explicitly enables CUDA inside its mocked toolkit fixture, so Windows CI's inherited CUDA-disabled CMake arguments cannot change the expected index. The no-ATen guard logs dependencies and checks library names, avoiding matches in paths or hexadecimal addresses while still rejecting Torch libraries and missing dependencies. The no-ATen job passes in CI.
Add PyTorch 2.15's
aoti_torch_dtype_element_sizeto the SlimTensor shims and regenerate the Windows import library. The library preserves all 51 existing exports, adds the new API, and is byte-for-byte reproducible across independent builds. A Windows COFF link probe fails with the old library and passes with the regenerated one; the compiled C ABI returns the correct sizes for all nine supported dtypes. All six model exports and the Windows CUDA runtime unit job pass in Windows CUDA CI.Restrict ROCm INT6 autotuning to eight-row tiles. An unmodified larger tile reproduced an illegal memory access for M=15, N=257, K=5376, while the reference agreed with a CPU calculation. All 576 remaining configuration and shape checks passed against that reference in the mitigation run. The normal INT6 launch, autotuning, dynamic-shape and dispatch tests also pass: 51 tests. This is a release mitigation with a possible performance tradeoff on ROCm; CUDA retains its existing candidates and the accuracy tolerances are unchanged.
Route tag-triggered runner refreshes through a workflow dispatch on protected main, preserving the runner-group environment's protected-branch policy. The current RC1 runner refresh has already been applied successfully from main: six groups updated, additions only.
Local validation also passed all 26 CUDA matrix tests with inherited
CMAKE_ARGS=-DEXECUTORCH_BUILD_CUDA=OFF, dependency-guard fixtures, workflow YAML parsing, formatting and whitespace checks. The broader test and Windows model execution runs are linked above. CoreML's timeout is outside this change.Authored with assistance from OpenAI Codex.