Move pull.yml to linux_job_v3

[huydhn]

Third of six splitting up #22107. Stacked on #22246.

45 call sites in one file, all mechanical: the v2 -> v3 rename, EC2 runner labels swapped for their OSDC equivalents, use-custom-docker-registry dropped since v3 ignores it, and the ECR image spelled out against the hash _docker-image.yml resolves.

EC2	OSDC
linux.2xlarge, linux.2xlarge.memory	mt-l-x86iavx512-8-64
linux.4xlarge.memory	mt-l-x86iavx512-16-128
linux.24xlarge	mt-l-x86iavx512-94-192
linux.24xlarge.memory	mt-l-x86iavx512-94-768
linux.arm64.2xlarge	mt-l-arm64g4-16-62
linux.g5.4xlarge.nvidia.gpu	mt-l-x86aavx2-29-113-a10g

Authored with Claude Code.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/22247

• 📄 Preview Python docs built from this PR

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

❌ 1 New Failure, 2 Unrelated Failures, 1 Unclassified Failure

As of commit 01e69e9 with merge base f665a3b ():

NEW FAILURE - The following job has failed:

• Test RISC-V Backend / test-riscv (yolo26, false, false) / run / linux-job (gh)
  RuntimeError: Command bash /__w/_temp/exec_script failed with exit code 1

UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:

• trunk / test-llama-runner-mac (fp32, coreml) / macos-job (gh) (this job did not run on the merge base, so DrCI cannot tell whether the failure is pre-existing)
  AttributeError: 'BackendConfig' object has no attribute 'mps'

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / test-moshi-linux / linux-job (gh) (matched linux rule in flaky-rules.json)
  E: Failed to fetch http://security.ubuntu.com/ubuntu/pool/main/libs/libssh/libssh-gcrypt-4_0.9.6-2ubuntu0.22.04.7_amd64.deb 404 Not Found [IP: 2620:2d:4002:1::103 80]

BROKEN TRUNK - The following job failed but was present on the merge base:

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

• Test RISC-V Backend / test-riscv (yolo26, true, false) / run / linux-job (gh) (trunk failure)
  AssertionError: Expected p_model_model_0_conv_weight to be NCHW in channels last reshape pass

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

[huydhn]

Last commit temporarily points pull.yml's 45 linux_job_v3 refs at pytorch/test-infra#8683 (osdc/hf-cache-writable) to exercise the HF cache mode fix. Reverts to @main before landing.

The job to watch is test-models-linux (ic4, portable, ...), which is one of the ones failing with Errno 30 Read-only file system on models--timm--inception_v4.tf_in1k.

Worth knowing what this run can and cannot prove: this PR triggers a pull_request run, where the label is already visible in the event payload, so it exercises the refresh path but not the ciflow-tag resolution #8683 adds. It also will not seed the shared cache — meta-pytorch/pytorch-gha-infra#1463 is still open, so the sync step fails assume-role and is skipped. So a pass here means "the refresh path downloads what it needs and the tests go green", not "the cache is fixed".

[huydhn]

Pushed the unittest-nxp-neutron fix, folded into the existing worker-cap commit rather than added on top, since it is the same fix reaching one more call site.

backends/nxp/run_unittests.sh runs pytest -c /dev/null -n "logical", and logical resolves through psutil.cpu_count() just like auto does — so it fanned out to the node's core count on an 8 vCPU pod and starved the aot_neutron_compile tests past their hardcoded 300s subprocess timeout. My original sweep grepped for auto only and missed this spelling.

Verified the variable governs logical too, despite its name:

-n logical, no env var           -> 16 workers
PYTEST_XDIST_AUTO_NUM_WORKERS=3  ->  3 workers

Re-grepped for both spellings across .ci, backends, examples and extension. The remaining hits are unittest-macos-cmake.sh (macOS, dedicated VMs, unaffected), a docstring in backends/test/suite/flows/nxp.py, and two READMEs — so all four executable call sites are now covered.

From the previous run on this branch, the two job families that were failing are green: test-models-linux across the whole matrix including ic4 (was Errno 30 Read-only file system), and all four test-arm-backend-no-driver cells on mt-l-x86iavx512-8-64 (was OOMKilled at 64 workers). The only other failure in that run was test-models-macos-cpu (resnet50), a 502 from gitlab.arm.com fetching KleidiAI — unrelated to the migration.