Repository navigation
Conversation
Use PyTorch 2.14.0 CPU and TorchAO 0.18.0 from release indexes for the MobileSAM Python 3.12 setup. Align FlatBuffers with TOSA to keep the example in one venv. Preserve GPU, source-build and Python 3.15 package selections. Let the shared FVP launcher qualify console measurements using its simulation mode, including runners without semihosting. Report unavailable CPU timing, ratios and fast-mode NPU counters once. Retain simulated NPU counters in timed mode and preserve hardware reporting. Remove the separate runner mode argument. Keep the README flow simple with an optional run.sh --timed. Validated the README install, export, cross-build and FVP flow in a fresh Python 3.12 venv on Ubuntu 24.04 amd64 in Docker under Apple silicon emulation: PASS, mask IoU 0.977785. Release-wheel hashes and output match the previously tested artifacts. Launcher regression tests cover both modes on U55, U65 and U85, valid zero counters, and error and execution-count preservation. Non-semihosting FVP runs pass in both modes with bundled output checks. Authored with assistance from Codex. Signed-off-by: Usamah Zaheer <usamah.zaheer@arm.com> Change-Id: I05452fe9a3a760673faf069cfd1110ef7d1d24f6
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24118
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 13 Unrelated FailuresAs of commit fe93a8b with merge base 0295293 ( NEW FAILURES - The following jobs have failed:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
Collaborator
|
@digantdesai or @rascani you probably want to check this one, do it need a internal test? |
Contributor
No internal test needed, but I think I'd like to better understand the problem with the torchao dependency. It seems like this has several conditionals that may just be working around temporary issues. |
This branch has not been deployed
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.
Summary
Fix the MobileSAM Python 3.12 CPU setup failing on an unavailable TorchAO nightly by using released PyTorch/TorchAO wheels and aligning FlatBuffers with TOSA's requirement.
Make the shared FVP launcher report unsupported CPU timing, CPU/NPU ratios and fast-mode NPU counters as unavailable. Timed runs retain simulated NPU counters, including valid zeros. This also covers runners built without semihosting and preserves hardware reporting. Keep the existing example flow and add
run.sh --timed.Test plan
./examples/arm/mobilesam_prompt_segmentation_example_ethos_u/run.shwith Python 3.12 on Ubuntu 24.04 amd64 in Docker under Apple silicon emulation: PASS, mask IoU 0.977785; output unchanged.python -m pytest -q .ci/scripts/tests/test_cu134_dependencies.py backends/arm/test/test_run_fvp.py: 30 tests passed.backends/arm/scripts/run_fvp.shwith and without--fast: bundled output checks passed; timed mode retained 3,127 NPU cycles and valid zero counters, while fast mode suppressed unavailable measurements.backends/arm/scripts/pre-push: all checks passed, including lint, commit/license checks, operator documentation and public API manifests.Authored with assistance from Codex.
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani