Support HunyuanImage-image3.0 for t2i#1250
Conversation
…n. now, this version support fa2/fa3, flashinfer. and we also enable cfg parallel for generation, which combines 3 modes (batch, serial, parallel)
…parallel, sequence parallel kv-all-gather, sequence parallel ulysses
…parallel, sequence parallel kv-all-gather, sequence parallel ulysses
…parallel, sequence parallel kv-all-gather, sequence parallel ulysses
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces several configuration files for Hunyuan Image3 text-to-image and image-to-image inference, including distributed parallel configurations, alongside an implementation plan for the 'Think Recaption' feature. Feedback on the configurations highlights that several distributed inference JSON files are missing the required "cfg_parallel" and/or "seq_parallel" boolean flags, which are necessary to actually enable the specified parallelization strategies.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| "pipeline_parallel": true, | ||
| "enable_cfg": true, |
There was a problem hiding this comment.
The configuration specifies "hunyuan_cfg_mode": "parallel" and a "cfg_p_size" of 2, but "cfg_parallel": true is missing. Without this flag, CFG parallel inference will not be enabled.
| "pipeline_parallel": true, | |
| "enable_cfg": true, | |
| "pipeline_parallel": true, | |
| "cfg_parallel": true, | |
| "enable_cfg": true, |
| "pipeline_parallel": true, | ||
| "hunyuan_image3_pipeline_layout": "interleaved", | ||
| "enable_cfg": true, |
There was a problem hiding this comment.
This configuration is intended for hybrid CFG and Sequence Parallel inference (using kv_all_gather), but both "cfg_parallel": true and "seq_parallel": true are missing. Without these flags, the parallel features will not be enabled.
| "pipeline_parallel": true, | |
| "hunyuan_image3_pipeline_layout": "interleaved", | |
| "enable_cfg": true, | |
| "pipeline_parallel": true, | |
| "cfg_parallel": true, | |
| "seq_parallel": true, | |
| "hunyuan_image3_pipeline_layout": "interleaved", | |
| "enable_cfg": true, |
| "pipeline_parallel": true, | ||
| "hunyuan_image3_pipeline_layout": "interleaved", | ||
| "enable_cfg": true, |
There was a problem hiding this comment.
This configuration is intended for hybrid CFG and Sequence Parallel inference (using ulysses), but both "cfg_parallel": true and "seq_parallel": true are missing. Without these flags, the parallel features will not be enabled.
| "pipeline_parallel": true, | |
| "hunyuan_image3_pipeline_layout": "interleaved", | |
| "enable_cfg": true, | |
| "pipeline_parallel": true, | |
| "cfg_parallel": true, | |
| "seq_parallel": true, | |
| "hunyuan_image3_pipeline_layout": "interleaved", | |
| "enable_cfg": true, |
| "pipeline_parallel": true, | ||
| "enable_cfg": true, |
There was a problem hiding this comment.
This configuration is intended for Sequence Parallel inference (using kv_all_gather), but "seq_parallel": true is missing. Without this flag, sequence parallel will not be enabled.
| "pipeline_parallel": true, | |
| "enable_cfg": true, | |
| "pipeline_parallel": true, | |
| "seq_parallel": true, | |
| "enable_cfg": true, |
| "pipeline_parallel": true, | ||
| "enable_cfg": true, |
There was a problem hiding this comment.
This configuration is intended for Sequence Parallel inference (using ulysses), but "seq_parallel": true is missing. Without this flag, sequence parallel will not be enabled.
| "pipeline_parallel": true, | |
| "enable_cfg": true, | |
| "pipeline_parallel": true, | |
| "seq_parallel": true, | |
| "enable_cfg": true, |
Chernobyllight
changed the title
Chernobyllight
changed the title
1 participant
this lightx2v version support HunyuanImage-image3.0 for t2i, which implements FA2/FA3, flashinfer, cfg parallel(batch, serial, parallel), sequence parallel(kv-all-gather sp, ulysses sp)