[model]Add UltraFlux-v1-image support by erfgss · Pull Request #611 · vllm-project/vllm-omni

erfgss · 2026-01-04T09:29:06Z

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

Purpose

Add UltraFlux-v1-image support #327

Test Plan

python text_to_image.py \
  --model Owen777/UltraFlux-v1 \
  --prompt "A vast rocky landscape dominated by towering, weathered stone formations, bathed in the ethereal glow of a vibrant night sky filled with a sea of stars, the Milky Way stretching across the heavens, captured from a low angle to emphasize the immense scale of the rocks against the expansive cosmos above. The scene is illuminated by soft, cool moonlight, casting long, dramatic shadows on the textured rock surfaces. The color palette is rich with deep blues, purples, and silvery whites, creating a serene, otherworldly atmosphere." \
  --height 4096 \
  --width 4096 \
  --output UltraFlux-v1_image_output.png \
  --cache_backend cache_dit

Test Result

without cache_dit

Processed prompts: 100%|██████████| 1/1 [02:00<00:00, 120.95s/img, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-04 03:05:42 [omni.py:687] [Summary] {'e2e_requests': 1,mg, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-04 03:05:42 [omni.py:687]  'e2e_total_time_ms': 120952.95739173889,
INFO 01-04 03:05:42 [omni.py:687]  'e2e_sum_time_ms': 120951.51662826538,
INFO 01-04 03:05:42 [omni.py:687]  'e2e_total_tokens': 0,
INFO 01-04 03:05:42 [omni.py:687]  'e2e_avg_time_per_request_ms': 120951.51662826538,
INFO 01-04 03:05:42 [omni.py:687]  'e2e_avg_tokens_per_s': 0.0,
INFO 01-04 03:05:42 [omni.py:687]  'wall_time_ms': 120952.95739173889,
INFO 01-04 03:05:42 [omni.py:687]  'final_stage_id': {'0_5208ad13-e972-40b0-b30a-45fbac8b7d4e': 0},
INFO 01-04 03:05:42 [omni.py:687]  'stages': [{'stage_id': 0,
INFO 01-04 03:05:42 [omni.py:687]              'requests': 1,
INFO 01-04 03:05:42 [omni.py:687]              'tokens': 0,
INFO 01-04 03:05:42 [omni.py:687]              'total_time_ms': 120951.89571380615,
INFO 01-04 03:05:42 [omni.py:687]              'avg_time_per_request_ms': 120951.89571380615,
INFO 01-04 03:05:42 [omni.py:687]              'avg_tokens_per_s': 0.0}],
INFO 01-04 03:05:42 [omni.py:687]  'transfers': []}
Adding requests:   0%|          | 0/1 [02:00<?, ?it/s]
[Stage-0] ERROR 01-04 03:05:42 [omni_stage.py:636] Received shutdown signal
[Stage-0] INFO 01-04 03:05:42 [gpu_worker.py:265] Worker 0: Received shutdown message
[Stage-0] INFO 01-04 03:05:42 [gpu_worker.py:287] event loop terminated.
[Stage-0] INFO 01-04 03:05:43 [gpu_worker.py:318] Worker 0: Shutdown complete.
INFO 01-04 03:05:46 [text_to_image.py:168] Outputs: [OmniRequestOutput(request_id='', finished=True, stage_id=0, final_output_type='image', request_output=[OmniRequestOutput(request_id='0_5208ad13-e972-40b0-b30a-45fbac8b7d4e', finished=True, stage_id=None, final_output_type='image', request_output=None, images=[1 PIL Images], prompt='A vast rocky landscape dominated by towering, weathered stone formations, bathed in the ethereal glow of a vibrant night sky filled with a sea of stars, the Milky Way stretching across the heavens, captured from a low angle to emphasize the immense scale of the rocks against the expansive cosmos above. The scene is illuminated by soft, cool moonlight, casting long, dramatic shadows on the textured rock surfaces. The color palette is rich with deep blues, purples, and silvery whites, creating a serene, otherworldly atmosphere.', latents=None, metrics={})], images=[], prompt=None, latents=None, metrics={})]
Saved generated image to UltraFlux-v1_image_output.png

with cache_dit

Processed prompts: 100%|██████████| 1/1 [00:43<00:00, 43.56s/img, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-04 03:09:29 [omni.py:687] [Summary] {'e2e_requests': 1,g, est. speed stage-0 img/s: 0.00, avg e2e_lat: 0.0ms]
INFO 01-04 03:09:29 [omni.py:687]  'e2e_total_time_ms': 43562.40963935852,
INFO 01-04 03:09:29 [omni.py:687]  'e2e_sum_time_ms': 43560.83941459656,
INFO 01-04 03:09:29 [omni.py:687]  'e2e_total_tokens': 0,
INFO 01-04 03:09:29 [omni.py:687]  'e2e_avg_time_per_request_ms': 43560.83941459656,
INFO 01-04 03:09:29 [omni.py:687]  'e2e_avg_tokens_per_s': 0.0,
INFO 01-04 03:09:29 [omni.py:687]  'wall_time_ms': 43562.40963935852,
INFO 01-04 03:09:29 [omni.py:687]  'final_stage_id': {'0_81384414-7d7f-4871-ace1-48322e07f1f2': 0},
INFO 01-04 03:09:29 [omni.py:687]  'stages': [{'stage_id': 0,
INFO 01-04 03:09:29 [omni.py:687]              'requests': 1,
INFO 01-04 03:09:29 [omni.py:687]              'tokens': 0,
INFO 01-04 03:09:29 [omni.py:687]              'total_time_ms': 43561.26618385315,
INFO 01-04 03:09:29 [omni.py:687]              'avg_time_per_request_ms': 43561.26618385315,
INFO 01-04 03:09:29 [omni.py:687]              'avg_tokens_per_s': 0.0}],
INFO 01-04 03:09:29 [omni.py:687]  'transfers': []}
Adding requests:   0%|          | 0/1 [00:43<?, ?it/s]
[Stage-0] ERROR 01-04 03:09:29 [omni_stage.py:636] Received shutdown signal
[Stage-0] INFO 01-04 03:09:29 [gpu_worker.py:265] Worker 0: Received shutdown message
[Stage-0] INFO 01-04 03:09:29 [gpu_worker.py:287] event loop terminated.
[Stage-0] INFO 01-04 03:09:29 [gpu_worker.py:318] Worker 0: Shutdown complete.
INFO 01-04 03:09:32 [text_to_image.py:168] Outputs: [OmniRequestOutput(request_id='', finished=True, stage_id=0, final_output_type='image', request_output=[OmniRequestOutput(request_id='0_81384414-7d7f-4871-ace1-48322e07f1f2', finished=True, stage_id=None, final_output_type='image', request_output=None, images=[1 PIL Images], prompt='A vast rocky landscape dominated by towering, weathered stone formations, bathed in the ethereal glow of a vibrant night sky filled with a sea of stars, the Milky Way stretching across the heavens, captured from a low angle to emphasize the immense scale of the rocks against the expansive cosmos above. The scene is illuminated by soft, cool moonlight, casting long, dramatic shadows on the textured rock surfaces. The color palette is rich with deep blues, purples, and silvery whites, creating a serene, otherworldly atmosphere.', latents=None, metrics={})], images=[], prompt=None, latents=None, metrics={})]
Saved generated image to UltraFlux-v1_image_output.png

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, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
(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: 7af3b2287a

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-01-04T09:34:34Z

vllm_omni/diffusion/registry.py

        "pipeline_wan2_2",
        "Wan22Pipeline",
    ),


Preserve existing diffusion model registrations

This registry block now omits several existing pipelines (e.g., WanImageToVideoPipeline, BagelPipeline, LongCatImageEditPipeline, StableDiffusion3Pipeline) even though their modules still exist under vllm_omni/diffusion/models. Since initialize_model looks up od_config.model_class_name in _DIFFUSION_MODELS and raises a ValueError when missing, any config that previously used those model_class_name values will now fail to initialize. Please restore those entries or explicitly deprecate them across docs/configs.

Useful? React with 👍 / 👎.

chatgpt-codex-connector · 2026-01-04T09:34:34Z

vllm_omni/diffusion/registry.py

+    #UltraFlux
+    "FluxPipeline": (
+        "ultraflux-v1_image",
+        "pipeline_ultraflux",
+        "UltraFluxPipeline",


Register UltraFlux under UltraFluxPipeline key

The new UltraFlux entry is keyed as FluxPipeline, but the actual class and docs use UltraFluxPipeline. If users set model_class_name=UltraFluxPipeline as documented, _DIFFUSION_MODELS lookup will fail and initialize_model will raise “Model class UltraFluxPipeline not found.” Please register it under UltraFluxPipeline (or update naming consistently) so the model is loadable.

Useful? React with 👍 / 👎.

david6666666 · 2026-01-04T09:37:01Z

vllm_omni/diffusion/cache/cache_dit_backend.py

    {
        "WanPipeline": enable_cache_for_wan22,
        "FluxPipeline": enable_cache_for_flux,
+        "UltraFluxPipeline": enable_cache_for_flux,


maybe you need to update new func name such as enable_cache_for_ultraflux

david6666666 · 2026-01-04T09:37:44Z

vllm_omni/diffusion/registry.py

        "pipeline_wan2_2",
        "Wan22Pipeline",
    ),
-    "WanImageToVideoPipeline": (


do not delete, just add your pipeline

david6666666 · 2026-01-04T09:37:58Z

vllm_omni/diffusion/registry.py

-        "StableDiffusion3Pipeline",
-    ),
+    #UltraFlux
+    "FluxPipeline": (


rename UltraFluxPipeline

david6666666 · 2026-01-04T09:38:09Z

vllm_omni/diffusion/registry.py

    od_config: OmniDiffusionConfig,
 ):
    model_class = DiffusionModelRegistry._try_load_model_cls(od_config.model_class_name)
+    print("DEBUG model_class =", model_class)


david6666666 · 2026-01-05T02:28:08Z

vllm_omni/diffusion/registry.py

        "LongCatImagePipeline",
    ),
-    "BagelPipeline": (
+        "BagelPipeline": (


revert here

david6666666 · 2026-01-05T02:29:09Z

vllm_omni/diffusion/registry.py

    # where mod_folder and mod_relname are  defined and mapped using `_DIFFUSION_MODELS` via the `arch` key
    "QwenImageEditPipeline": "get_qwen_image_edit_pre_process_func",
    "QwenImageEditPlusPipeline": "get_qwen_image_edit_plus_pre_process_func",
+    "QwenImageLayeredPipeline": "get_qwen_image_layered_pre_process_func",


remove this

erfgss · 2026-01-06T03:03:42Z

@SamitHuang

hsliuustc0106 · 2026-01-12T11:21:53Z

vllm_omni/diffusion/diffusion_engine.py

@@ -4,7 +4,7 @@
 import multiprocessing as mp


adding a model should not change diffusion_engine

hsliuustc0106 · 2026-01-12T11:25:08Z

docs/models/supported_models.md

 | `WanPipeline` | Wan2.2-T2V, Wan2.2-TI2V | `Wan-AI/Wan2.2-T2V-A14B-Diffusers`, `Wan-AI/Wan2.2-TI2V-5B-Diffusers` |
 | `WanImageToVideoPipeline` | Wan2.2-I2V | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` |
 | `OvisImagePipeline` | Ovis-Image | `OvisAI/Ovis-Image` |
 |`LongcatImagePipeline` | LongCat-Image | `meituan-longcat/LongCat-Image` |


how about https://docs.vllm.ai/projects/vllm-omni/en/latest/user_guide/diffusion_acceleration/

hsliuustc0106 · 2026-01-12T11:26:13Z

mkdocs.yml

        - "vllm_omni.entrypoints.async_diffusion"  # avoid importing vllm in mkdocs building
        - "vllm_omni.entrypoints.openai"  # avoid importing vllm in mkdocs building
        - "vllm_omni.entrypoints.openai.protocol"  # avoid importing vllm in mkdocs building
+        - "vllm_omni.entrypoints.omni"  # avoid importing vllm in mkdocs building


why we need to change this?

hsliuustc0106 · 2026-01-12T11:26:43Z

vllm_omni/entrypoints/async_omni.py


            # Summarize and print stats
            try:
+                import json as _json


why we need to change this? @Bounty-hunter PTAL

hsliuustc0106 · 2026-01-12T11:27:34Z

vllm_omni/entrypoints/log_utils.py

@@ -4,6 +4,7 @@
 from dataclasses import dataclass


this PR is designed for adding a model, you should not make any changes to these comment files

hsliuustc0106 · 2026-01-12T11:31:44Z

vllm_omni/entrypoints/omni.py

    def _initialize_stages(self, model: str, kwargs: dict[str, Any]) -> None:
        """Initialize stage list management."""
-        stage_init_timeout = kwargs.get("stage_init_timeout", 20)
+        # Diffusion/large models can take long to load; align default with CLI (300s)


you should not change this default value, you need to provide this in your cli

hsliuustc0106 · 2026-01-12T11:32:24Z

vllm_omni/entrypoints/omni_stage.py

                pass
    logger.debug("Engine initialized")

+    # Check if stage engine supports profiling (via vLLM's built-in profiler)


why you need change omni_stage in the model support PR?

hsliuustc0106 · 2026-01-12T11:33:23Z

vllm_omni/outputs.py

                }
            )

        return result


should not change in this PR

hsliuustc0106 · 2026-01-16T10:43:34Z

vllm_omni/diffusion/registry.py

-        "StableDiffusion3Pipeline",
-    ),
+    #UltraFlux
+    "FluxPipeline": (


hsliuustc0106 · 2026-01-16T10:44:31Z

vllm_omni/diffusion/models/ultraflux-v1_image/transformer_flux.py

@@ -0,0 +1,931 @@
+# Copyright 2025 Black Forest Labs, The HuggingFace Team and The InstantX Team. All rights reserved.


add support to TP please check #735

hsliuustc0106 · 2026-01-16T10:45:03Z

docs/models/supported_models.md

 | `WanImageToVideoPipeline` | Wan2.2-I2V | `Wan-AI/Wan2.2-I2V-A14B-Diffusers` |
 | `OvisImagePipeline` | Ovis-Image | `OvisAI/Ovis-Image` |
 |`LongcatImagePipeline` | LongCat-Image | `meituan-longcat/LongCat-Image` |
 |`LongCatImageEditPipeline` | LongCat-Image-Edit | `meituan-longcat/LongCat-Image-Edit` |


please also update diffusion acceleration md for cache dit support

hsliuustc0106 · 2026-01-16T10:46:13Z

@david6666666 can we not use benchmark serving under benchmarks folder for t2i jobs

david6666666 · 2026-01-19T02:55:54Z

Please make similar modifications based on the review comments in #809.

please use attention layer in vllm_omni/diffusion.
we have rope layer in vllm-omni.
Just import to reduce copying local functions from.
use from vllm.model_executor.layers.layernorm import RMSNorm
use from vllm.use from vllm.model_executor.layers.linear import QKVParallelLinear, ReplicatedLinear

Signed-off-by: Chen Yang <[email protected]>

…in Ring Attention (vllm-project#767) Signed-off-by: XU Mingshi <[email protected]> Signed-off-by: mxuax <[email protected]> Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: Sihyeon Jang <[email protected]> Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: zjy0516 <[email protected]> Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: iwzbi <[email protected]> Signed-off-by: Chen Yang <[email protected]>

…#722) Signed-off-by: ZeldaHuang <[email protected]> Signed-off-by: Chen Yang <[email protected]>

…ject#781) Signed-off-by: Yuhan Liu <[email protected]> Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: David Chen <[email protected]> Signed-off-by: Chen Yang <[email protected]> # Conflicts: # vllm_omni/diffusion/registry.py

Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: erfgss <[email protected]>

Signed-off-by: Chen Yang <[email protected]>

erfgss · 2026-01-19T07:31:17Z

Please make similar modifications based on the review comments in #809.

please use attention layer in vllm_omni/diffusion.

we have rope layer in vllm-omni.

Just import to reduce copying local functions from.

use from vllm.model_executor.layers.layernorm import RMSNorm

use from vllm.use from vllm.model_executor.layers.linear import QKVParallelLinear, ReplicatedLinear

erfgss · 2026-01-19T07:37:34Z

Please make similar modifications based on the review comments in #809.

please use attention layer in vllm_omni/diffusion.

we have rope layer in vllm-omni.

Just import to reduce copying local functions from.

use from vllm.model_executor.layers.layernorm import RMSNorm

use from vllm.use from vllm.model_executor.layers.linear import QKVParallelLinear, ReplicatedLinear

Signed-off-by: Chen Yang <[email protected]>

Signed-off-by: erfgss <[email protected]>

david6666666 · 2026-02-04T02:59:07Z

Support SP(ulysses, ring)
Support TP
Support CFG parallel
validate Cache-DiT

Signed-off-by: erfgss <[email protected]>

lishunyang12

Thanks for the contribution — left a few comments inline on things I noticed.

lishunyang12 · 2026-02-22T01:39:16Z

docs/models/supported_models.md

-|`Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-VoiceDesign | `Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign` |
-|`Qwen3TTSForConditionalGeneration` | Qwen3-TTS-12Hz-1.7B-Base | `Qwen/Qwen3-TTS-12Hz-0.6B-Base` |
-
+|`UltraFluxPipeline` | UltraFlux-v1 | `Owen777/UltraFlux-v1` |


Looks like this diff might have accidentally removed the three Qwen3-TTS entries — probably a rebase artifact? The UltraFlux line should be added alongside them.

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/pipeline_ultraflux.py

+        self.default_sample_size = 64
+
+        print(self.vae.config)
+        print(self.transformer.config)


A few debug print() calls here — might want to swap them for logger.debug() or remove before merging.

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/pipeline_ultraflux.py

+            scheduler=scheduler,
+        )
+
+        self.vae_scale_factor = 32


Quick question — vae_scale_factor = 32 while standard Flux uses 16. Is 32 correct for UltraFlux, or a copy-paste from somewhere? If it's intentional, a brief comment explaining why would be helpful.

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/transformer_flux.py

+def _get_fused_projections(attn: "FluxAttention", hidden_states, encoder_hidden_states=None):
+    query, key, value = attn.to_qkv(hidden_states).chunk(3, dim=-1)
+
+    encoder_query = encoder_key = encoder_value = (None,)


I think there might be a small issue here — (None,) creates a 1-tuple rather than assigning None to all three variables. This could cause problems downstream when calling .unflatten(...) on a tuple. Maybe:

encoder_query = encoder_key = encoder_value = None

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/transformer_flux.py

+            mscale = torch.where(
+                scale <= 1.0, torch.tensor(1.0, device=scale.device, dtype=scale.dtype), 0.1 * torch.log(scale) + 1.0
+            )
+            mscale = torch.where(


Looks like this mscale computation is a duplicate of lines 612-614. Probably a copy-paste leftover — the second one could be removed.

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/transformer_flux.py

+        self.added_kv_proj_dim = added_kv_proj_dim
+        self.added_proj_bias = added_proj_bias
+
+        self.norm_q = torch.nn.RMSNorm(dim_head, eps=eps, elementwise_affine=elementwise_affine)


I saw in the earlier review that the maintainer suggested using from vllm.model_executor.layers.layernorm import RMSNorm instead of torch.nn.RMSNorm. Looks like it still needs to be updated here and on lines 300, 311, 312.

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/pipeline_ultraflux.py

+        for tok_name in ("tokenizer", "tokenizer_2"):
+            tok = getattr(self.pipe, tok_name, None)
+            if tok is not None and hasattr(tok, "model_max_length"):
+                tok.model_max_length = 512


Just wondering — hardcoding tok.model_max_length = 512 overrides whatever the tokenizer originally had. Would it make sense to read this from the model config instead?

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/cache/cache_dit_backend.py

-        "LongCatImagePipeline": enable_cache_for_longcat_image,
-        "LongCatImageEditPipeline": enable_cache_for_longcat_image,
+        "UltraFluxPipeline": enable_cache_for_ultraflux,
+        "LongcatImagePipeline": enable_cache_for_longcat_image,


The LongCatImagePipeline -> LongcatImagePipeline rename seems like a separate change. Might be worth mentioning in the PR description, or splitting it out if you prefer?

lishunyang12 · 2026-02-22T01:39:16Z

vllm_omni/diffusion/models/ultraflux-v1_image/pipeline_ultraflux.py

+        self.tokenizer_max_length = (
+            self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
+        )
+        self.default_sample_size = 64


With default_sample_size = 64 and vae_scale_factor = 32, the default resolution would be 2048x2048, but the test plan uses 4096x4096. Is 2048 the intended default?

erfgss requested a review from hsliuustc0106 as a code owner January 4, 2026 09:29

chatgpt-codex-connector bot reviewed Jan 4, 2026

View reviewed changes

david6666666 requested changes Jan 4, 2026

View reviewed changes

david6666666 reviewed Jan 5, 2026

View reviewed changes

david6666666 mentioned this pull request Jan 5, 2026

[RFC]: DiT model and feature support enhancement #85

Closed

58 tasks

hsliuustc0106 reviewed Jan 12, 2026

View reviewed changes

erfgss force-pushed the feat/UltraFlux-v1-image branch from 560c1a1 to 448588b Compare January 13, 2026 08:54

david6666666 mentioned this pull request Jan 16, 2026

vLLM-Omni Model Support #808

Open

57 tasks

hsliuustc0106 reviewed Jan 16, 2026

View reviewed changes

erfgss and others added 18 commits January 19, 2026 10:58

feat/UltraFlux-v1-image

7a69c19

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

0818688

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

d117835

Signed-off-by: Chen Yang <[email protected]>

[Bugfix] Raise ValueError when joint_strategy='rear' and causal=True …

159fba2

…in Ring Attention (vllm-project#767) Signed-off-by: XU Mingshi <[email protected]> Signed-off-by: mxuax <[email protected]> Signed-off-by: Chen Yang <[email protected]>

[Feat] add vllm-omni version collection (vllm-project#740)

f130723

Signed-off-by: Sihyeon Jang <[email protected]> Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

060bb48

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

cc50071

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

793ad0d

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

098473c

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

789930c

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

059ec13

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

3adc8db

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

bb0618f

Signed-off-by: Chen Yang <[email protected]>

[Doc] refactor diffusion doc (vllm-project#753)

074cb93

Signed-off-by: zjy0516 <[email protected]> Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

cc733a1

Signed-off-by: Chen Yang <[email protected]>

[Bugfix] Fix stable diffusion3 compatibility error (vllm-project#772)

31c245f

Signed-off-by: iwzbi <[email protected]> Signed-off-by: Chen Yang <[email protected]>

[Feature] Support Qwen3 Omni talker mtp batch inference (vllm-project…

2bab2c0

…#722) Signed-off-by: ZeldaHuang <[email protected]> Signed-off-by: Chen Yang <[email protected]>

[BugFix]Remove duplicate error handling for request results (vllm-pro…

35b5a7a

…ject#781) Signed-off-by: Yuhan Liu <[email protected]> Signed-off-by: Chen Yang <[email protected]>

erfgss added 5 commits January 19, 2026 11:07

feat/UltraFlux-v1-image

bf5d081

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

5042c97

Signed-off-by: Chen Yang <[email protected]>

[Model] add flux2 klein (vllm-project#809)

e45994c

Signed-off-by: David Chen <[email protected]> Signed-off-by: Chen Yang <[email protected]> # Conflicts: # vllm_omni/diffusion/registry.py

feat/UltraFlux-v1-image

b04bf98

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

67a48fb

Signed-off-by: Chen Yang <[email protected]>

erfgss force-pushed the feat/UltraFlux-v1-image branch from 8fc9d02 to 67a48fb Compare January 19, 2026 03:18

erfgss and others added 10 commits January 19, 2026 11:18

Merge branch 'main' into feat/UltraFlux-v1-image

47fd0c6

Signed-off-by: erfgss <[email protected]>

feat/UltraFlux-v1-image

3cc5cfb

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

21e2132

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

f626c02

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

8ee88ae

Signed-off-by: Chen Yang <[email protected]>

Merge branch 'main' into feat/UltraFlux-v1-image

efacd5e

Merge branch 'main' into feat/UltraFlux-v1-image

c0c5ace

feat/UltraFlux-v1-image

c088962

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

32fe65d

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

f085fe3

Signed-off-by: Chen Yang <[email protected]>

erfgss added 4 commits January 19, 2026 15:43

feat/UltraFlux-v1-image

e643b8b

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

781a16a

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

f58a207

Signed-off-by: Chen Yang <[email protected]>

feat/UltraFlux-v1-image

43eab3a

Signed-off-by: Chen Yang <[email protected]>

erfgss requested a review from david6666666 January 19, 2026 08:18

erfgss added 3 commits January 22, 2026 10:44

Merge branch 'main' into feat/UltraFlux-v1-image

d8a81e5

Merge branch 'main' into feat/UltraFlux-v1-image

7346450

Signed-off-by: erfgss <[email protected]>

Merge branch 'main' into feat/UltraFlux-v1-image

7ac98e2

erfgss added 2 commits February 5, 2026 09:11

Merge branch 'main' into feat/UltraFlux-v1-image

d24ca44

Merge branch 'main' into feat/UltraFlux-v1-image

3d8dfb3

Signed-off-by: erfgss <[email protected]>

lishunyang12 reviewed Feb 22, 2026

View reviewed changes

		@@ -0,0 +1,931 @@
		# Copyright 2025 Black Forest Labs, The HuggingFace Team and The InstantX Team. All rights reserved.

Comments

Conversation

erfgss commented Jan 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan

Test Result

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Jan 4, 2026

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector bot Jan 4, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

erfgss commented Jan 6, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

hsliuustc0106 commented Jan 16, 2026

Uh oh!

david6666666 commented Jan 19, 2026

Uh oh!

erfgss commented Jan 19, 2026

Uh oh!

erfgss commented Jan 19, 2026

Uh oh!

david6666666 commented Feb 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lishunyang12 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lishunyang12 Feb 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lishunyang12 Feb 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

erfgss commented Jan 4, 2026 •

edited

Loading

david6666666 commented Feb 4, 2026 •

edited

Loading

lishunyang12 left a comment •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading

lishunyang12 Feb 22, 2026 •

edited

Loading