[Bugfix] Allow to enable HSDP alone

[gcanlin]

PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTTOM) HAVE BEEN CONSIDERED.

Purpose

For standalone HSDP: when all other parallelism dimensions are 1, but fully_shard_degree > 1, use fully_shard_degree as dit_parallel_size. This ensures orthogonal rank generation works correctly for HSDP workers

Test Plan

vllm serve Wan-AI/Wan2.2-TI2V-5B-Diffusers --omni --port 8091  --use-hsdp --hsdp-shard-size 4

Test Result

[Stage-0] INFO 02-28 08:29:42 [diffusers_loader.py:301] Loading weights took 2.36 seconds
[Stage-0] INFO 02-28 08:29:42 [hsdp.py:128] HSDP Inference: replicate_size=1, shard_size=4, world_size=4, rank=0, fs_world_size=4, fs_rank=0
[Stage-0] INFO 02-28 08:29:42 [diffusers_loader.py:301] Loading weights took 2.17 seconds
[Stage-0] INFO 02-28 08:29:42 [hsdp.py:128] HSDP Inference: replicate_size=1, shard_size=4, world_size=4, rank=2, fs_world_size=4, fs_rank=2
[Stage-0] INFO 02-28 08:29:42 [diffusers_loader.py:301] Loading weights took 2.32 seconds
[Stage-0] INFO 02-28 08:29:42 [hsdp.py:128] HSDP Inference: replicate_size=1, shard_size=4, world_size=4, rank=1, fs_world_size=4, fs_rank=1
[Stage-0] INFO 02-28 08:29:42 [diffusers_loader.py:301] Loading weights took 2.20 seconds
[Stage-0] INFO 02-28 08:29:42 [hsdp.py:128] HSDP Inference: replicate_size=1, shard_size=4, world_size=4, rank=3, fs_world_size=4, fs_rank=3

---

Essential Elements of an Effective PR Description Checklist

• The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
• The test plan. Please provide the test scripts & test commands. Please state the reasons if your codes don't require additional test scripts. For test file guidelines, please check the test style doc
• The test results. Please paste the results comparison before and after, or the e2e results.
• (Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model. Please run mkdocs serve to sync the documentation editions to ./docs.
• (Optional) Release notes update. If your change is user-facing, please update the release notes draft.

BEFORE SUBMITTING, PLEASE READ https://github.com/vllm-project/vllm-omni/blob/main/CONTRIBUTING.md (anything written below this line will be removed by GitHub Actions)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 74fbf6fa6f

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[hsliuustc0106]

Missing regression test for standalone HSDP mode. Consider adding a unit test that verifies is_standalone_hsdp detection and dit_parallel_size calculation with various parallel configs (standalone HSDP, HSDP+SP, HSDP+TP) to prevent future regressions.

[gcanlin]

Thanks for reminding! Done now.