Skip to content

[Fix] HunyuanImage-3.0: unify naming hunyuan_image_3 → hunyuan_image3#2712

Merged
Gaohan123 merged 1 commit intovllm-project:mainfrom
TaffyOfficial:fix/hunyuan-image3-rename
Apr 15, 2026
Merged

[Fix] HunyuanImage-3.0: unify naming hunyuan_image_3 → hunyuan_image3#2712
Gaohan123 merged 1 commit intovllm-project:mainfrom
TaffyOfficial:fix/hunyuan-image3-rename

Conversation

@TaffyOfficial
Copy link
Copy Markdown
Contributor

Purpose

Fix inconsistent naming hunyuan_image_3hunyuan_image3 across the codebase.

The upstream HF repo (tencent/HunyuanImage-3.0-Instruct) uses hunyuan_image_3 (underscore before "3") in file names and imports, but
vllm-omni's convention omits it (e.g. glm_image, qwen2_5_omni). The main branch currently mixes both spellings: directories use
hunyuan_image_3 while registry keys, stage config filenames, and test names already use hunyuan_image3. This causes 3 runtime
crashes:

  1. diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py:77
  2. diffusion/models/hunyuan_image3/hunyuan_image3_tokenizer.py:16
  3. platforms/musa/platform.py:41

All fail with: ModuleNotFoundError: No module named 'vllm_omni.diffusion.models.hunyuan_image_3'

Changes:

  • Rename diffusion/models/hunyuan_image_3/hunyuan_image3/
  • Rename stage configs: hunyuan_image3_moe_dit.yamlhunyuan_image3_t2i.yaml
  • Delete obsolete hunyuan_image_3_moe.yaml
  • Fix import path strings in registry.py, interface.py, musa/platform.py
  • Update test_hunyuanimage3_text2img.py to reference renamed config

Note: NPU/MUSA platform changes are string-only path fixes. The original paths were already broken at runtime. No NPU/MUSA hardware
available for testing.

Test Plan

This is a pure rename + string fix. No new test scripts needed — the existing test_hunyuanimage3_text2img.py has been updated to
reference the renamed config path.

Verify no remaining broken references:

grep -rn "hunyuan_image_3" vllm_omni/diffusion/ vllm_omni/platforms/ vllm_omni/model_executor/stage_configs/

Test Result

$ grep -rn "hunyuan_image_3" vllm_omni/diffusion/ vllm_omni/platforms/ vllm_omni/model_executor/stage_configs/
(no output — all references unified)

- The purpose of the PR
- The test plan
- The test results

@chatgpt-codex-connector
Copy link
Copy Markdown

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.

Copy link
Copy Markdown
Collaborator

@hsliuustc0106 hsliuustc0106 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Two config files still use the old hunyuan_image_3 naming on this branch:

  1. vllm_omni/model_executor/stage_configs/hunyuan_image_3_moe_2gpu.yaml
  2. vllm_omni/platforms/xpu/stage_configs/hunyuan_image_3_moe.yaml

Could you rename these for consistency?

@TaffyOfficial TaffyOfficial force-pushed the fix/hunyuan-image3-rename branch 2 times, most recently from ebf2d20 to 84d9137 Compare April 13, 2026 11:46
@TaffyOfficial
Copy link
Copy Markdown
Contributor Author

@hsliuustc0106 update

Copy link
Copy Markdown
Collaborator

@Gaohan123 Gaohan123 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please globaly check if other scripts and docs are also modified

@Gaohan123 Gaohan123 added the ready label to trigger buildkite CI label Apr 15, 2026
The upstream HF repo (tencent/HunyuanImage-3.0-Instruct) uses
`hunyuan_image_3` (underscore before "3") in file names and imports,
but vllm-omni's convention omits it (e.g. `glm_image`, `qwen2_5_omni`).
The main branch currently mixes both spellings: directories use
`hunyuan_image_3` while registry keys, stage config filenames, and test
names already use `hunyuan_image3`. This causes 3 runtime crashes:

1. diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py:77
2. diffusion/models/hunyuan_image3/hunyuan_image3_tokenizer.py:16
3. platforms/musa/platform.py:41

All fail with: ModuleNotFoundError: No module named
'vllm_omni.diffusion.models.hunyuan_image_3'

This commit:
- Renames diffusion/models/hunyuan_image_3/ → hunyuan_image3/
- Renames stage configs: hunyuan_image3_moe_dit.yaml → hunyuan_image3_t2i.yaml
- Deletes obsolete hunyuan_image_3_moe.yaml
- Fixes import path strings in registry.py, interface.py, musa/platform.py
- Updates test_hunyuanimage3_text2img.py to reference renamed config

Note: NPU/MUSA platform changes are string-only path fixes. The original
paths were already broken at runtime. No NPU/MUSA hardware available for
testing.

Signed-off-by: TaffyOfficial <2324465096@qq.com>
@TaffyOfficial TaffyOfficial force-pushed the fix/hunyuan-image3-rename branch from 84d9137 to 615f902 Compare April 15, 2026 04:21
Copy link
Copy Markdown
Collaborator

@Gaohan123 Gaohan123 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@Gaohan123 Gaohan123 enabled auto-merge (squash) April 15, 2026 04:35
@Gaohan123 Gaohan123 merged commit 1ad726f into vllm-project:main Apr 15, 2026
7 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ready label to trigger buildkite CI

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants