[BREAKING][model, recipe, tests, doc] chore: remove v0.3.0 deprecated Qwen3-Omni legacy path - #522
Open
zhtmike wants to merge 1 commit into
Open
Conversation
… Qwen3-Omni legacy path The legacy Qwen3-Omni monkey-patch package and its verl.trainer.main_ppo launchers were deprecated when the V1 trainer (verl_omni.trainer.main_omni) landed, with removal scheduled for v0.3.0 (see verl-project#359, verl-project#312). v0.2.0 is published, so the scheduled removals are now executed: - delete verl_omni/models/transformers/ (legacy thinker monkey-patches) - delete the deprecated main_ppo launchers, their stage/recipe configs, and the legacy e2e smoke script + config - drop the deleted module from the ci-e2e-omni gpu-smoke trigger patterns - update examples/gspo_trainer/README.md and docs/start/models.md to point NPU users at run_qwen3_omni_thinker_gspo_npu_avqa_v1.sh AI assistance (ZCode) was used for this change. Co-authored-by: ZCode Signed-off-by: Mike Cheung <zhtmike@gmail.com>
zhtmike
requested review from
AndyZhou952,
SamitHuang,
knlnguyen1802 and
wtomin
as code owners
September 3, 2026 08:32
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
The GPU smoke-test docs still list qwen3_omni_thinker.py as a ci-e2e-omni trigger even though it was removed from the GROUP_PATTERNS mapping.
Once you've addressed the issues Copilot identified, you can request another Copilot review.
Pull request overview
Removes the pre–v0.3.0 legacy Qwen3-Omni “external_lib monkey-patch” path (the verl_omni/models/transformers/ patches), along with the deprecated verl.trainer.main_ppo-based example launchers/configs and the legacy e2e smoke script, keeping the V1 trainer (verl_omni.trainer.main_omni) as the supported path.
Changes:
- Delete the legacy Qwen3-Omni transformer monkey-patch module/package.
- Remove deprecated v0 launcher scripts + associated stage/recipe YAMLs and the legacy e2e smoke test.
- Update smoke-group selection patterns and user docs to drop the removed legacy paths.
File summaries
| File | Description |
|---|---|
| verl_omni/models/transformers/qwen3_omni_thinker.py | Remove deprecated Qwen3-Omni monkey-patch module. |
| verl_omni/models/transformers/init.py | Remove deprecated transformers patch package init (package removal). |
| tests/special_e2e/run_gspo_qwen3_omni_thinker_lora_smoke.sh | Remove legacy main_ppo + external_lib-based e2e smoke script. |
| tests/special_e2e/qwen3_omni_thinker_only_smoke.yaml | Remove legacy vLLM-Omni stage config used only by the deprecated smoke path. |
| tests/gpu_smoke/select_gpu_smoke_groups.py | Update ci-e2e-omni trigger patterns to drop removed legacy patch file. |
| examples/gspo_trainer/README.md | Update examples docs to remove references to deprecated legacy launchers and clarify V1 path. |
| examples/gspo_trainer/qwen3_omni/run_qwen3_omni_thinker_gspo_npu.sh | Remove deprecated NPU full-parameter legacy launcher. |
| examples/gspo_trainer/qwen3_omni/run_qwen3_omni_thinker_gspo_lora.sh | Remove deprecated GPU LoRA legacy launcher. |
| examples/gspo_trainer/qwen3_omni/qwen3_omni_thinker_only.yaml | Remove legacy stage config used only by the deprecated launcher path. |
| examples/gspo_trainer/qwen3_omni/qwen3_omni_thinker_only_npu.yaml | Remove legacy NPU stage config used only by the deprecated launcher path. |
| examples/gspo_trainer/qwen3_omni/config/qwen3_omni_thinker_gspo.yaml | Remove deprecated recipe YAML for the legacy main_ppo path. |
| docs/start/models.md | Remove deprecated “GSPO (full, NPU)” legacy script entry from supported trainers table. |
Review details
- Files reviewed: 12/12 changed files
- Comments generated: 1
- Review effort level: Lite
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
This was referenced Sep 3, 2026
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.
What does this PR do?
Executes the removals scheduled before v0.3.0: the legacy Qwen3-Omni monkey-patch
package (
verl_omni/models/transformers/), the deprecatedverl.trainer.main_ppolaunchers and their stage/recipe configs, and the legacye2e smoke script. The V1 trainer (
verl_omni.trainer.main_omni) path isunaffected. Deprecation notices were introduced in #359 and #312; v0.2.0 is
published, so the scheduled removals are now due.
Not duplicating any open PR: searched for deprecation/removal PRs; the only
related open PR is #509 (Wan2.2 DanceGRPO v0 docs), a different scope.
BREAKING: removes the launchers
run_qwen3_omni_thinker_gspo_lora.sh/run_qwen3_omni_thinker_gspo_npu.shand their configs. NPU users should move torun_qwen3_omni_thinker_gspo_npu_avqa_v1.sh.Test
Checklist Before Submitting
examples/gspo_trainer/README.md,docs/start/models.md).AI Assistance
AI assistance (ZCode) was used for implementation and PR preparation. The human
submitter reviewed every changed line and ran the tests above.