Metal CI: Add Qwen 3.5 MoE tiny model integration test#19042
Metal CI: Add Qwen 3.5 MoE tiny model integration test#19042manuelcandales merged 1 commit intomainfrom
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/19042
Note: Links to docs will display an error until the docs builds have been completed. ❗ 1 Active SEVsThere are 1 currently active SEVs. If your PR is affected, please view them below: ❌ 3 New Failures, 1 Cancelled Job, 6 Pending, 4 Unrelated FailuresAs of commit 2330a5a with merge base 6be4fb5 ( NEW FAILURES - The following jobs have failed:
CANCELLED JOB - The following job was cancelled. Please retry:
FLAKY - The following job failed but was likely due to flakiness present on trunk:
BROKEN TRUNK - The following jobs failed but was 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. |
This PR needs a
|
There was a problem hiding this comment.
Pull request overview
Adds a new Metal CI integration test job that exports and runs the Qwen 3.5 MoE tiny model on macOS Metal, aiming to validate both prefill and decode execution paths.
Changes:
- Extend Metal workflow PR path filters to include the Qwen3.5 MoE example and LLM export code.
- Add a new
test-metal-qwen35-moe-tinyjob that installs ExecuTorch, exports a tiny Qwen3.5 MoE model for Metal, builds the runner, generates a minimal tokenizer, and runs two inference scenarios.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Export the tiny model with --backend metal, build the C++ runner, and verify both decode (T=1) and prefill (T>2) complete successfully. Uses a byte-level tokenizer matching the tiny model's vocab_size=256. Authored with Claude.
e40ab21 to
2330a5a
Compare
Export the tiny model with --backend metal, build the C++ runner, and verify both decode (T=1) and prefill (T>2) complete successfully. Uses a byte-level tokenizer matching the tiny model's vocab_size=256.
Authored with Claude.