A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 01:45 UTC
[Docs] Note image preprocessing difference between qwen_vl_utils and vllm. (#43393)
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io> Signed-off-by: wang.yuqi <noooop@126.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
wang.yuqi committed
May 22, 2026 at 16:43 UTC
2380bfc2104267914eea36015e2a347b9318c6c0
3 files changed
+10
-6
docs/models/pooling_models/embed.md
+4
-1
@@ -91,7 +91,7 @@ You can compute pairwise similarity scores to build a similarity matrix using th
91
| `LlamaNemotronVLModel` | Llama Nemotron Embedding + SigLIP | T + I | `nvidia/llama-nemotron-embed-vl-1b-v2` | | |
92
| `LlavaNextForConditionalGeneration`<sup>C</sup> | LLaVA-NeXT-based | T / I | `royokong/e5-v` | | ✅︎ |
93
| `Phi3VForCausalLM`<sup>C</sup> | Phi-3-Vision-based | T + I | `TIGER-Lab/VLM2Vec-Full` | | ✅︎ |
94
-| `Qwen3VLForConditionalGeneration`<sup>C</sup> | Qwen3-VL | T + I + V | `Qwen/Qwen3-VL-Embedding-2B`, etc. | ✅︎ | ✅︎ |
94
+| `Qwen3VLForConditionalGeneration`<sup>C</sup> (see note) | Qwen3-VL | T + I + V | `Qwen/Qwen3-VL-Embedding-2B`, etc. | ✅︎ | ✅︎ |
95
| `SiglipModel` | SigLIP, SigLIP2 | T / I | `google/siglip-base-patch16-224`, `google/siglip2-base-patch16-224` | | |
96
| `*ForConditionalGeneration`<sup>C</sup>, `*ForCausalLM`<sup>C</sup>, etc. | Generative models | \* | N/A | \* | \* |
97
@@ -102,6 +102,9 @@ If your model is not in the above list, we will try to automatically convert the
102
[as_embedding_model][vllm.model_executor.models.adapters.as_embedding_model]. By default, the embeddings
103
of the whole prompt are extracted from the normalized hidden state corresponding to the last token.
104
105
+!!! note
106
+ `Qwen3-VL-Embedding` officially uses `qwen_vl_utils` for image preprocessing, while vLLM uses `transformers`' `video_processing_qwen3_vl`, which leads to slightly different results compared to the official Hugging Face repository examples. Example code for offline inference using `qwen_vl_utils` can be found in the [vision_embedding_offline.py](../../../examples/pooling/embed/vision_embedding_offline.py) example.
107
+
108
!!! note
109
Although vLLM supports automatically converting models of any architecture into embedding models via --convert embed, to get the best results, you should use pooling models that are specifically trained as such.
110
docs/models/pooling_models/scoring.md
+1
-1
@@ -103,7 +103,7 @@ The three supported scoring functions are as illustrated in the figure below.
103
\* Feature support is the same as that of the original model.
104
105
!!! note
106
- Similar to Qwen3-Reranker, you need to use the following `--hf_overrides` to load the official original `Qwen3-VL-Reranker`.
106
+ Similar to Qwen3-Reranker, you need to use the following `--hf_overrides` to load the official original `Qwen3-VL-Reranker`. `Qwen3-VL` officially uses `qwen_vl_utils` for image preprocessing, while vLLM uses `transformers`' `video_processing_qwen3_vl`, which leads to slightly different results compared to the official Hugging Face repository examples.
107
108
```bash
109
vllm serve Qwen/Qwen3-VL-Reranker-2B --hf_overrides '{"architectures": ["Qwen3VLForSequenceClassification"],"classifier_from_token": ["no", "yes"],"is_original_qwen3_reranker": true}'
docs/models/supported_models.md
+5
-4
@@ -619,13 +619,13 @@ These models primarily accept the [`LLM.generate`](./generative_models.md#llmgen
619
| `QianfanOCRForConditionalGeneration` | QianfanOCR | T + I<sup>E+</sup> | `baidu/Qianfan-OCR`, etc. | ✅︎ | ✅︎ |
620
| `QwenVLForConditionalGeneration`<sup>^</sup> | Qwen-VL | T + I<sup>E+</sup> | `Qwen/Qwen-VL`, `Qwen/Qwen-VL-Chat`, etc. | ✅︎ | ✅︎ |
621
| `Qwen2AudioForConditionalGeneration` | Qwen2-Audio | T + A<sup>+</sup> | `Qwen/Qwen2-Audio-7B-Instruct` | | ✅︎ |
622
-| `Qwen2VLForConditionalGeneration` | QVQ, Qwen2-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/QVQ-72B-Preview`, `Qwen/Qwen2-VL-7B-Instruct`, `Qwen/Qwen2-VL-72B-Instruct`, etc. | ✅︎ | ✅︎ |
623
-| `Qwen2_5_VLForConditionalGeneration` | Qwen2.5-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen2.5-VL-3B-Instruct`, `Qwen/Qwen2.5-VL-72B-Instruct`, etc. | ✅︎ | ✅︎ |
622
+| `Qwen2VLForConditionalGeneration` <sup>Q</sup> | QVQ, Qwen2-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/QVQ-72B-Preview`, `Qwen/Qwen2-VL-7B-Instruct`, `Qwen/Qwen2-VL-72B-Instruct`, etc. | ✅︎ | ✅︎ |
623
+| `Qwen2_5_VLForConditionalGeneration` <sup>Q</sup> | Qwen2.5-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen2.5-VL-3B-Instruct`, `Qwen/Qwen2.5-VL-72B-Instruct`, etc. | ✅︎ | ✅︎ |
624
| `Qwen2_5OmniThinkerForConditionalGeneration` | Qwen2.5-Omni | T + I<sup>E+</sup> + V<sup>E+</sup> + A<sup>+</sup> | `Qwen/Qwen2.5-Omni-3B`, `Qwen/Qwen2.5-Omni-7B` | ✅︎ | ✅︎ |
625
| `Qwen3_5ForConditionalGeneration` | Qwen3.5 | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3.5-9B-Instruct`, etc. | ✅︎ | ✅︎ |
626
| `Qwen3_5MoeForConditionalGeneration` | Qwen3.5-MOE | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3.5-35B-A3B-Instruct`, etc. | ✅︎ | ✅︎ |
627
-| `Qwen3VLForConditionalGeneration` | Qwen3-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3-VL-4B-Instruct`, etc. | ✅︎ | ✅︎ |
628
-| `Qwen3VLMoeForConditionalGeneration` | Qwen3-VL-MOE | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3-VL-30B-A3B-Instruct`, etc. | ✅︎ | ✅︎ |
627
+| `Qwen3VLForConditionalGeneration` <sup>Q</sup> | Qwen3-VL | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3-VL-4B-Instruct`, etc. | ✅︎ | ✅︎ |
628
+| `Qwen3VLMoeForConditionalGeneration` <sup>Q</sup> | Qwen3-VL-MOE | T + I<sup>E+</sup> + V<sup>E+</sup> | `Qwen/Qwen3-VL-30B-A3B-Instruct`, etc. | ✅︎ | ✅︎ |
629
| `Qwen3OmniMoeThinkerForConditionalGeneration` | Qwen3-Omni | T + I<sup>E+</sup> + V<sup>E+</sup> + A<sup>+</sup> | `Qwen/Qwen3-Omni-30B-A3B-Instruct`, `Qwen/Qwen3-Omni-30B-A3B-Thinking` | ✅︎ | ✅︎ |
630
| `Qwen3ASRForConditionalGeneration` | Qwen3-ASR | T + A<sup>+</sup> | `Qwen/Qwen3-ASR-1.7B` | ✅︎ | ✅︎ |
631
| `RForConditionalGeneration` | R-VL-4B | T + I<sup>E+</sup> | `YannQi/R-4B` | | ✅︎ |
@@ -647,6 +647,7 @@ Some models are supported only via the [Transformers modeling backend](#transfor
647
<sup>E</sup> Pre-computed embeddings can be inputted for this modality.</br>
648
<sup>+</sup> Multiple items can be inputted per text prompt for this modality.
649
<sup>*</sup> Only specific variants of the model support this modality (see notes below).</br>
650
+<sup>Q</sup> `Qwen*-VL` officially uses `qwen_vl_utils` for image preprocessing, while vLLM uses `transformers`' `video_processing_qwen*`, which leads to slightly different results compared to the official Hugging Face repository examples.
651
652
!!! note
653
`Gemma3nForConditionalGeneration` is only supported on V1 due to shared KV caching and it depends on `timm>=1.0.17` to make use of its