[ZeroRedundancyOptimizer] Pytorch compliant state

[blefaudeux]

Same as #52760 which I could not get to land. I just could not live with ghstack/ghimport/randomly broken things, I break enough of them myself, so this is a fresh copy without ghstack shenanigans. I'm hopeful that this can land relatively bug free, and am sorry for the duplications..

What this does:

• call the common_utils test runner instead of unittest, because it seems that it's how it should be done
• change the returned state from ZeroRedundancyOptimizer to be PyTorch compliant, which has the added benefit of being elastic (world size independent)

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 98cc2f3 (more details on the Dr. CI page):

---

• 1/1 failures introduced in this PR

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pytorch_linux_xenial_py3_clang7_onnx_ort_test2 (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

Feb 27 05:46:44 E AssertionError: False is not true

Feb 27 05:46:44         size = workspace.FetchBlob(bsize_map[blobs[0]])
Feb 27 05:46:44         self.assertEqual(tuple(), size.shape)
Feb 27 05:46:44         self.assertEqual(min(num_to_collect, max_example_to_cover), size.item())
Feb 27 05:46:44     
Feb 27 05:46:44         hist, _ = np.histogram(
Feb 27 05:46:44             reference_result[:, 0], bins=10, range=(1, max_example_to_cover)
Feb 27 05:46:44         )
Feb 27 05:46:44         print("Sample histogram: {}".format(hist))
Feb 27 05:46:44     
Feb 27 05:46:44 >       self.assertTrue(all(hist > 0.6 * (num_to_collect / 10)))
Feb 27 05:46:44 E       AssertionError: False is not true
Feb 27 05:46:44 
Feb 27 05:46:44 /opt/conda/lib/python3.6/site-packages/caffe2/python/operator_test/dataset_ops_test.py:690: AssertionError
Feb 27 05:46:44 ----------------------------- Captured stdout call -----------------------------
Feb 27 05:46:44 Collect Net Proto: name: "collect_net_10"
Feb 27 05:46:44 op {
Feb 27 05:46:44   input: "blob_1_vec"
Feb 27 05:46:44   input: "blob_2_vec"
Feb 27 05:46:44   input: "blob_3_vec"
Feb 27 05:46:44   input: "blob_1"
Feb 27 05:46:44   input: "blob_2"

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

[blefaudeux]

The broken unit test seems unrelated (see https://app.circleci.com/pipelines/github/pytorch/pytorch/278895/workflows/f139e3a0-f43d-4fcb-a45d-c84799ce1d84/jobs/11199323)

[mrshenli]

Already reviewed on #52760.

@blefaudeux would I be correct if I assume the additional lines compared to #52760 are code format changes and the logic is still the same?

[mrshenli]

The broken unit test seems unrelated

the error indeed look irrelevant. I think it should be safe to land.

Error: No such container:path: fa36bb4f7a10af04bbab8e7159921ade2e68270d129326355f9e4132e21b4be9:/var/lib/jenkins/workspace/test/.coverage

[blefaudeux]

Already reviewed on #52760.

@blefaudeux would I be correct if I assume the additional lines compared to #52760 are code format changes and the logic is still the same?

Yes, I'm not sure how the sequence unrolled but it's a simple copy of #52760 except for the unit test runner baked in (so basically 52760 and prior)

[facebook-github-bot]

@blefaudeux has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

@blefaudeux merged this pull request in 249c213.