[Bugfix] Make mrope kwargs optional in HunyuanImage3 get_mrope_input_positions

[ianliuy]

Purpose

Fix #2642 HunyuanImage3ForConditionalGeneration.get_mrope_input_positions() crashes with TypeError: missing 3 required keyword-only arguments: 'hf_config', 'image_grid_thw', and 'video_grid_thw' when called from the upstream vllm caller.

Root cause

The SupportsMRoPE protocol defines get_mrope_input_positions(self, input_tokens, mm_features) only 2 positional args. But HunyuanImage3's implementation declares hf_config, image_grid_thw, and video_grid_thw as required keyword-only parameters (after *, no defaults). The upstream vllm caller follows the protocol and only passes 2 args crash.

Key finding

These 3 kwargs are never referenced in the ~180-line method body. The method extracts all needed data from self.config and mm_features internally (e.g., vae_token_grid_hw and vit_spatial_shapes from mm_feature.data). They were likely carried over from an older Qwen2-VL style signature.

Fix

Give the 3 unused kwargs None defaults so both callers work:

• Upstream vllm (passes 2 positional args) now works
• vllm-omni override (passes all kwargs explicitly) still works

- hf_config: PretrainedConfig,
- image_grid_thw: list[list[int]] | torch.Tensor,
- video_grid_thw: list[list[int]] | torch.Tensor,
+ hf_config: PretrainedConfig | None = None,
+ image_grid_thw: list[list[int]] | torch.Tensor | None = None,
+ video_grid_thw: list[list[int]] | torch.Tensor | None = None,

Test Plan

Signature-only change no new logic paths introduced. The 3 kwargs are confirmed unused in the method body (verified by reading all ~180 lines). Both callers (upstream 2-arg and vllm-omni full-kwargs) continue to work correctly.

Test Result

Before: TypeError: get_mrope_input_positions() missing 3 required keyword-only arguments
After: Method accepts calls with or without these kwargs.

---

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.
• 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.
• (Optional) Release notes update.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[ianliuy]

@hsliuustc0106 @lishunyang12 The readthedocs check passed on the original commit (65f076f) but failed on the merge commit (eba1ebc) with a build timeout (15m limit). Since this PR only modifies a single Python file and no documentation, the timeout was likely introduced by recent changes in main that were pulled in via "Update branch". Could we re-trigger the readthedocs build, or bypass it for merge?

[ianliuy]

@hsliuustc0106 @lishunyang12 Gentle ping the readthedocs timeout on the merge commit appears to be a repo-wide issue (not specific to this PR). All other checks pass. Could you take a look when convenient?