testintel: add missing test for AvgDown3D.forward (created by linxwang) by jiagaoxiang · Pull Request #1096 · AMD-AGI/Primus

jiagaoxiang · 2026-09-04T16:28:02Z

Same-repo refile of #1083 so Primus-CI-TAS runs against AMD-AGI/Primus (fork PRs do not receive Docker Hub credentials, so build-docker / torch unit tests never ran).

This PR was created by linxwang via Test Gap Resolver.

No existing test in tests/unit_tests/backends/diffusion (or anywhere else in the repo) instantiates AvgDown3D or imports vae2_2.py at all; the pad->view->permute->view->view->mean sequence that implements grouped spatiotemporal average-downsampling, including the temporal padding branch (pad_t>0) and the group_size averaging for group_size>1 with in_channels != out_channels (plus the independent spatial/channel mapping oracle), is completely unexercised.

This PR adds a focused unit test for AvgDown3D.forward; it does not change production code.

Candidate cand:9cf47afdeacb6b6d0378129eb4bf01c2 at 2f01706.

…rence Address review: compare full output against a hand-computed reference for factor_s=2 with in_channels != out_channels (deterministic torch.arange input), verifying spatial block grouping and channel/group mapping instead of only asserting output shape.

pre-commit's black hook wraps the AvgDown3D(...) constructor call in test_forward_downsamples_spatial_and_channel_dims onto multiple lines (line was too long). No behavior change.

Copilot

🟢 Approval recommended

The change is isolated to a new unit test file and the assertions align with the current AvgDown3D implementation behavior.

Pull request overview

Adds targeted unit-test coverage for the WAN VAE AvgDown3D downsampling module to ensure its grouped channel averaging and temporal/spatial downsampling behaviors are exercised in CI.

Changes:

Introduces a new unit test module that instantiates AvgDown3D and validates output shape/value semantics.
Covers both the temporal left-padding branch (pad_t > 0) and the grouped-channel averaging path (group_size behavior).
Adds a negative test to confirm incompatible (in_channels * factor) % out_channels != 0 is rejected.

File summaries

File	Description
tests/unit_tests/backends/diffusion/test_wan_vae_avgdown3d.py	New focused tests validating `AvgDown3D.forward` behavior (padding, reshape/permute mapping, averaging) plus init assertion.

Review details

Files reviewed: 1/1 changed files
Comments generated: 0
Review effort level: Lite

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

WangLingxun · 2026-09-07T03:33:18Z

The revised test itself looks good and passes against current main when overlaid locally. One small PR-description correction: in_channels * factor != out_channels * group_size cannot be the covered condition because group_size is defined so those quantities are equal. The non-trivial case covered here is group_size > 1 with in_channels != out_channels, plus the independent spatial/channel mapping oracle. Please adjust that sentence; no test-code change is needed.

jiagaoxiang · 2026-09-07T03:35:59Z

TestIntel PR Steward: Thanks @WangLingxun for the catch. You're right that in_channels * factor and out_channels * group_size are equal by construction of group_size, so that condition can't be what's covered. I've updated the PR description to say the non-trivial case exercised here is group_size > 1 with in_channels != out_channels, plus the independent spatial/channel mapping oracle. No test-code change was needed, per your note.

Copilot's automated review came back with an approval recommendation and no actionable comments. This PR remains draft; I'll keep watching for CI, further review comments, and a human approval before merging.

WangLingxun · 2026-09-07T04:20:33Z

@@ -0,0 +1,94 @@
+###############################################################################
+# Copyright (c) 2025, Advanced Micro Devices, Inc.


This is a newly added file, so the copyright year should be 2026.

Address review comment from WangLingxun: newly added file should use 2026.

jiagaoxiang · 2026-09-07T04:26:01Z

TestIntel PR Steward: Addressing @WangLingxun's review comment on tests/unit_tests/backends/diffusion/test_wan_vae_avgdown3d.py line 2: updated the copyright header from 2025 to 2026 since this is a newly added file. Pushed as commit 25bde0a (new head SHA 25bde0a06ee3ba09785b41a2003d1e3d7841e1bb).

No other outstanding review comments. Still draft; watching for CI, further review, and a human approval before requesting a merge.

Copilot

🟢 Approval recommended

The change is test-only, aligns with the stated goal, and the assertions match the current AvgDown3D implementation (padding and reshape/permute/group-mean behavior).

Review details

Files reviewed: 1/1 changed files
Comments generated: 0 new
Review effort level: Lite

…b4bf01c2

Copilot

Copilot was unable to review this pull request because the user who requested the review is ineligible. To be eligible to request a review, you need a paid Copilot license, or your organization must enable Copilot code review.

jiagaoxiang · 2026-09-07T06:47:21Z

TestIntel PR Steward: Investigated the CI failure on the current head (f6f662c, @WangLingxun's merge of main into this branch).

Most of the checks the poll flagged as failed (build-docker, code-lint, dependency-review, run-unittest-jax) belong to an earlier, superseded workflow run (34089417005) that GitHub cancelled when the merge commit triggered a fresh run — those are stale cancellations, not real failures.

The one genuine failure, in the latest run (34089420822, run-unittest-torch), is tests/trainer/test_megatron_trainer.py::TestProjectionSimulate::test_memory_benchmark erroring with:

torch.distributed.DistNetworkError: The server socket has failed to listen on any local network address. port: 53395, ... EADDRINUSE, message: address already in use

That's an unrelated test (Megatron trainer memory benchmark, not AvgDown3D) hitting a leftover/occupied rendezvous port on the runner — a classic infra flake, not a defect introduced by this PR's single new test file. I've requeued the failed jobs on that run and will watch for green.

…b4bf01c2

jiagaoxiang · 2026-09-07T09:53:45Z

TestIntel PR Steward: Status check on this poll:

CI: the failures the poll flagged (build-docker, code-lint, dependency-review, run-unittest-jax, run-unittest-torch) were all on the previous head f6f662c and are already resolved — the requeued attempt (run 34089420822, attempt 2) finished green across all jobs including run-unittest-torch, well before this poll.
Review: @WangLingxun's copyright-year comment was already addressed (commit 25bde0a) and the PR description was corrected per their follow-up note; no other outstanding review threads.
Approval: @Xiaoming-AMD approved.
Mergeable state was behind — two PRs landed on main (Scientific-notation CLI overrides (e.g. --lr 1e-5) are parsed as strings, not floats #1110, fix(config): parse scientific-notation CLI overrides as floats (e.g. --lr 1e-5) #1112 fixes and test: pin Qwen3 TurboAttention layout and qk-norm (PRPUNDIT-21) #1104) right around the time of approval, so this branch was out of date. I updated the branch from main (no content changes; new merge commit only). New head is 9844593 (was f6f662c).

CI is now running on the new head via run 34108505438. I'll merge once that run is green, provided the approval still stands and no one asks to hold.

Copilot

🟢 Approval recommended

It adds well-scoped, deterministic unit tests that match the implementation’s padding and reshape/permute semantics without changing production behavior.

Review details

Files reviewed: 1/1 changed files
Comments generated: 0 new
Review effort level: Lite

jiagaoxiang · 2026-09-07T11:56:08Z

TestIntel PR Steward: Merged (squash, 2f87a4b).

All conditions were met at this poll:

CI: all 6 jobs on Primus-CI-TAS run 34108505438 green on the current head (9844593) — dependency-review, code-lint, build-docker, run-unittest-jax, run-unittest-torch, coverage-summary — plus the Read the Docs check succeeded.
Review: @Xiaoming-AMD's human approval stands; @WangLingxun's copyright-year comment was addressed in 25bde0a and their PR-description correction was applied; no outstanding human change requests or hold/WIP comments.
Mergeable state: clean, no conflicts.

Thanks all for the review.

TestIntel and others added 3 commits September 3, 2026 01:38

testintel: add missing test for AvgDown3D.forward

f4394eb

Candidate cand:9cf47afdeacb6b6d0378129eb4bf01c2 at 2f01706.

style: apply black formatting to satisfy code-lint (3.12)

323b861

pre-commit's black hook wraps the AvgDown3D(...) constructor call in test_forward_downsamples_spatial_and_channel_dims onto multiple lines (line was too long). No behavior change.

Copilot AI lite review requested due to automatic review settings September 4, 2026 16:28

jiagaoxiang mentioned this pull request Sep 4, 2026

testintel: add missing test for AvgDown3D.forward (created by linxwang) #1083

Closed

Copilot started reviewing on behalf of jiagaoxiang September 4, 2026 16:29 View session

Copilot AI reviewed Sep 4, 2026

View reviewed changes

WangLingxun reviewed Sep 7, 2026

View reviewed changes

test: fix copyright year in new AvgDown3D test file

25bde0a

Address review comment from WangLingxun: newly added file should use 2026.

Copilot AI review requested due to automatic review settings September 7, 2026 04:25

Copilot started reviewing on behalf of jiagaoxiang September 7, 2026 04:26 View session

Copilot AI reviewed Sep 7, 2026

View reviewed changes

Merge branch 'main' into testintel/test/cand-9cf47afdeacb6b6d0378129e…

f6f662c

…b4bf01c2

Copilot AI review requested due to automatic review settings September 7, 2026 06:06

Copilot AI reviewed Sep 7, 2026

WangLingxun marked this pull request as ready for review September 7, 2026 06:07

WangLingxun requested review from Xiaoming-AMD, limou102 and wenxie-amd as code owners September 7, 2026 06:07

Xiaoming-AMD approved these changes Sep 7, 2026

View reviewed changes

Merge branch 'main' into testintel/test/cand-9cf47afdeacb6b6d0378129e…

9844593

…b4bf01c2

Copilot AI review requested due to automatic review settings September 7, 2026 09:53

Copilot started reviewing on behalf of jiagaoxiang September 7, 2026 09:54 View session

Copilot AI reviewed Sep 7, 2026

View reviewed changes

jiagaoxiang merged commit 2f87a4b into main Sep 7, 2026
10 checks passed

Navigation Menu

Uh oh!

testintel: add missing test for AvgDown3D.forward (created by linxwang) - #1096

testintel: add missing test for AvgDown3D.forward (created by linxwang)#1096
jiagaoxiang merged 6 commits into
mainfrom
testintel/test/cand-9cf47afdeacb6b6d0378129eb4bf01c2

jiagaoxiang commented Sep 4, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

WangLingxun commented Sep 7, 2026

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

WangLingxun Sep 7, 2026

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

		@@ -0,0 +1,94 @@
		###############################################################################
		# Copyright (c) 2025, Advanced Micro Devices, Inc.

Uh oh!

Conversation

jiagaoxiang commented Sep 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

WangLingxun commented Sep 7, 2026

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

WangLingxun Sep 7, 2026

Choose a reason for hiding this comment

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟢 Approval recommended

Uh oh!

Uh oh!

jiagaoxiang commented Sep 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

jiagaoxiang commented Sep 4, 2026 •

edited

Loading