Skip to content

GitLab

Commit 3d735feb authored by luopl's avatar luopl
Browse files

"Initial commit"

parents
Pipeline #3074 canceled with stages
Hide whitespace changes
Inline Side-by-side
Showing with 1550 additions and 0 deletions
LICENSE 0 → 100644
View file @ 3d735feb
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
README.md 0 → 100644
View file @ 3d735feb
# Qwen3-VL
## 论文
[Qwen3-VL Technical Report](https://arxiv.org/pdf/2511.21631)
## 模型简介
Qwen3-VL——迄今为止 Qwen 系列中最强大的视觉语言模型。
这一代产品在各方面都进行了全面升级：更出色的文本理解和生成能力、更深入的视觉感知和推理能力、更长的上下文长度、更强大的空间和视频动态理解能力以及更强大的代理交互能力。
提供从边缘到云端可扩展的密集型和 MoE 架构，以及可灵活按需部署的指导型和推理增强型思维型版本。
主要改进：
- 视觉代理：操作 PC/移动 GUI——识别元素、理解功能、调用工具、完成任务。
Visual Coding Boost：从图像/视频生成 Draw.io/HTML/CSS/JS。
- 高级空间感知：判断物体位置、视角和遮挡；提供更强的 2D 基础，并为空间推理和具身人工智能实现 3D 基础。
- 长上下文和视频理解：原生支持 256K 上下文，可扩展至 1M；能够处理书籍和数小时的视频，并具有完全回忆和二级索引功能。
- 增强型多模态推理能力：在 STEM/数学领域表现出色——因果分析和基于逻辑、证据的答案。
- 升级的视觉识别：更广泛、更高质量的预训练能够“识别一切”——名人、动漫、产品、地标、动植物等。
- 扩展的 OCR：支持 32 种语言（从 10 种增加到 32 种）；在弱光、模糊和倾斜条件下表现更佳；对罕见/古代字符和术语的处理能力更强；改进了长文档结构解析。
- 文本理解能力与纯 LLM 相当：实现无损、统一的文本视觉无缝融合，从而获得统一的理解。
模型架构更新：
<div align=center>
<img src="./doc/qwen3vl_arc.jpg"/>
</div>
1. **Interleaved-MRoPE**: 通过稳健的位置嵌入，在时间、宽度和高度上进行全频分配，增强长时程视频推理。
2. **DeepStack**: 融合多级 ViT 特征，以捕捉精细细节并增强图像与文本的对齐。
3. **Text–Timestamp Alignment:** 超越 T-RoPE，实现基于时间戳的精确事件定位，从而增强视频时间建模能力。
## 环境依赖
| 软件 | 版本 |
|:------------:| :------: |
| DTK | 25.04.2 |
| python | 3.10.12 |
| transformers | 4.57.1 |
| torch | 2.5.1+das.opt1.dtk25042 |
| accelerate | 1.11.0 |
| torchvision | 0.20.1+das.opt1.dtk25042 |
| flash_attn | 2.6.1+das.opt1.dtk2504|
| av | 16.0.1|
推荐使用镜像:
- 挂载地址`-v`，`{docker_name}`和 `{docker_image_name}`根据实际模型情况修改
```bash
docker run -it --shm-size 200g --network=host --name {docker_name} --privileged --device=/dev/kfd --device=/dev/dri --device=/dev/mkfd --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -u root -v /path/your_code_path/:/path/your_code_path/ -v /opt/hyhal/:/opt/hyhal/:ro {docker_image_name} bash
示例如下：
docker run -it --shm-size 200g --network=host --name qwen3vl --privileged --device=/dev/kfd --device=/dev/dri --device=/dev/mkfd --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -u root -v /path/your_code_path/:/path/your_code_path/ -v /opt/hyhal/:/opt/hyhal/:ro image.sourcefind.cn:5000/dcu/admin/base/vllm:0.9.2-ubuntu22.04-dtk25.04.2-py3.10 bash
#视频推理时安装PyAV后端依赖
pip install av
```
更多镜像可前往[光源](https://sourcefind.cn/#/service-list)下载使用。
关于本项目DCU显卡所需的特殊深度学习库可从[光合](https://developer.sourcefind.cn/tool/)开发者社区下载安装，其它包参照requirements.txt安装：
```
pip install -r requirements.txt
```
## 数据集
`暂无`
## 训练
`暂无`
## 推理
### transformers
#### 单机推理
```bash
#普通图文对话
HIP_VISIBLE_DEVICES=0 python qwen3vl_infer.py
#多图像推理
HIP_VISIBLE_DEVICES=0 python qwen3vl_infer_multi_images.py
#视频推理
HIP_VISIBLE_DEVICES=0 python qwen3vl_infer_video.py
```
## 效果展示
**场景1** ：普通图文对话
Input:
- image:
<div align=center>
<img src="./doc/demo.jpeg"/>
</div>
- text: "Describe this image."
Output:
<div align=center>
<img src="./doc/result.png"/>
</div>
**场景2** ：多图像推理
Input:
- image1:
<div align=center>
<img src="./doc/demo.jpeg"/>
</div>
- image2:
<div align=center>
<img src="./doc/dog.jpg"/>
</div>
- text: "Identify the similarities between these images."
Output:
<div align=center>
<img src="./doc/result_multi_images.png"/>
</div>
**场景3** ：视频推理
- Vedio:
![space_woaudio](./doc/space_woaudio.mp4)
- text:: "Describe this video."
Output:
<div align=center>
<img src="./doc/result_vedio.png"/>
</div>
### 精度
`DCU与GPU精度一致，推理框架：transformers。`
## 预训练权重
| 模型名称 | 权重大小 | DCU型号 | 最低卡数需求 |下载地址|
|:--------------------:|:----:|:----------:|:------:|:----------:|
| Qwen3-VL-4B-Instruct | 4B | BW1000| 1 | [Hugging Face](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct) |
| Qwen3-VL-8B-Instruct | 8B | BW1000| 1 | [Hugging Face](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct) |
## 源码仓库及问题反馈
- https://developer.sourcefind.cn/codes/modelzoo/qwen3-vl_pytorch
## 参考资料
- https://github.com/QwenLM/Qwen3-VL
\ No newline at end of file
README_ori.md 0 → 100644
View file @ 3d735feb
# Qwen3-VL
<p align="center">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vllogo.png" width="400"/>
<p>
<p align="center">
💜 <a href="https://chat.qwenlm.ai/"><b>Qwen Chat</b></a>&nbsp&nbsp | &nbsp&nbsp🤗 <a href="https://huggingface.co/collections/Qwen/qwen3-vl-68d2a7c1b8a8afce4ebd2dbe">Hugging Face</a>&nbsp&nbsp | &nbsp&nbsp🤖 <a href="https://modelscope.cn/collections/Qwen3-VL-5c7a94c8cb144b">ModelScope</a>&nbsp&nbsp | &nbsp&nbsp📑 <a href="https://qwen.ai/blog?id=99f0335c4ad9ff6153e517418d48535ab6d8afef&from=research.latest-advancements-list">Blog</a>&nbsp&nbsp | &nbsp&nbsp📚 <a href="https://github.com/QwenLM/Qwen3-VL/tree/main/cookbooks">Cookbooks</a>&nbsp&nbsp | &nbsp&nbsp📑 <a href="https://arxiv.org/pdf/2511.21631">Paper</a>&nbsp&nbsp
<br>
🖥️ <a href="https://huggingface.co/spaces/Qwen/Qwen3-VL-Demo">Demo</a>&nbsp&nbsp | &nbsp&nbsp💬 <a href="https://github.com/QwenLM/Qwen/blob/main/assets/wechat.png">WeChat (微信)</a>&nbsp&nbsp | &nbsp&nbsp🫨 <a href="https://discord.gg/CV4E9rpNSD">Discord</a>&nbsp&nbsp | &nbsp&nbsp📑 <a href="https://help.aliyun.com/zh/model-studio/developer-reference/qwen-vl-api">API</a>&nbsp&nbsp | &nbsp&nbsp🖥️ <a href="https://gallery.pai-ml.com/#/preview/deepLearning/cv/qwen2.5-vl">PAI-DSW</a>
</p>
## Introduction
Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.
This generation delivers comprehensive upgrades across the board: superior text understanding & generation, deeper visual perception & reasoning, extended context length, enhanced spatial and video dynamics comprehension, and stronger agent interaction capabilities.
Available in Dense and MoE architectures that scale from edge to cloud, with Instruct and reasoning‑enhanced Thinking editions for flexible, on‑demand deployment.
#### Key Enhancements:
* **Visual Agent**: Operates PC/mobile GUIs—recognizes elements, understands functions, invokes tools, completes tasks.
* **Visual Coding Boost**: Generates Draw.io/HTML/CSS/JS from images/videos.
* **Advanced Spatial Perception**: Judges object positions, viewpoints, and occlusions; provides stronger 2D grounding and enables 3D grounding for spatial reasoning and embodied AI.
* **Long Context & Video Understanding**: Native 256K context, expandable to 1M; handles books and hours-long video with full recall and second-level indexing.
* **Enhanced Multimodal Reasoning**: Excels in STEM/Math—causal analysis and logical, evidence-based answers.
* **Upgraded Visual Recognition**: Broader, higher-quality pretraining is able to “recognize everything”—celebrities, anime, products, landmarks, flora/fauna, etc.
* **Expanded OCR**: Supports 32 languages (up from 10); robust in low light, blur, and tilt; better with rare/ancient characters and jargon; improved long-document structure parsing.
* **Text Understanding on par with pure LLMs**: Seamless text–vision fusion for lossless, unified comprehension.
#### Model Architecture Updates:
<p align="center">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_arc.jpg" width="80%"/>
<p>
1. **Interleaved-MRoPE**: Full‑frequency allocation over time, width, and height via robust positional embeddings, enhancing long‑horizon video reasoning.
2. **DeepStack**: Fuses multi‑level ViT features to capture fine‑grained details and sharpen image–text alignment.
3. **Text–Timestamp Alignment:** Moves beyond T‑RoPE to precise, timestamp‑grounded event localization for stronger video temporal modeling.
## News
* 2025.11.27: We have released the [**Qwen3-VL paper**](https://arxiv.org/pdf/2511.21631), which introduces many technical details about Qwen3-VL, and we hope it will be helpful to everyone.
* 2025.10.21: We have released the **Qwen3-VL-2B** ([Instruct](https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct)/[Thinking](https://huggingface.co/Qwen/Qwen3-VL-2B-Thinking)) and **Qwen3-VL-32B** ([Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct)/[Thinking](https://huggingface.co/Qwen/Qwen3-VL-32B-Thinking)). Enjoy it!
* 2025.10.15: We have released the **Qwen3-VL-4B** ([Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)/[Thinking](https://huggingface.co/Qwen/Qwen3-VL-4B-Thinking)) and **Qwen3-VL-8B** ([Instruct](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct)/[Thinking](https://huggingface.co/Qwen/Qwen3-VL-8B-Thinking)). Enjoy it!
* 2025.10.4: We have released the [Qwen3-VL-30B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct) and [Qwen3-VL-30B-A3B-Thinking](https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Thinking). We have also released the FP8 version of the Qwen3-VL models — available in our [HuggingFace collection](https://huggingface.co/collections/Qwen/qwen3-vl-68d2a7c1b8a8afce4ebd2dbe) and [ModelScope collection](https://modelscope.cn/collections/Qwen3-VL-5c7a94c8cb144b).
* 2025.09.23: We have released the [Qwen3-VL-235B-A22B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct) and [Qwen3-VL-235B-A22B-Thinking](https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking). For more details, please check our [blog](https://qwen.ai/blog?id=99f0335c4ad9ff6153e517418d48535ab6d8afef&from=research.latest-advancements-list)!
* 2025.04.08: We provide the [code](https://github.com/QwenLM/Qwen2.5-VL/tree/main/qwen-vl-finetune) for fine-tuning Qwen2-VL and Qwen2.5-VL.
* 2025.03.25: We have released the [Qwen2.5-VL-32B](https://huggingface.co/Qwen/Qwen2.5-VL-32B-Instruct). It is smarter and its responses align more closely with human preferences. For more details, please check our [blog](https://qwenlm.github.io/blog/qwen2.5-vl-32b/)!
* 2025.02.20: we have released the [Qwen2.5-VL Technical Report](https://arxiv.org/abs/2502.13923). Alongside the report, we have also released AWQ-quantized models for Qwen2.5-VL in three different sizes: [3B](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct-AWQ), [7B](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct-AWQ) , and [72B](https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct-AWQ) parameters.
* 2025.01.28: We have released the [Qwen2.5-VL series](https://huggingface.co/Qwen). For more details, please check our [blog](https://qwenlm.github.io/blog/qwen2.5-vl/)!
* 2024.12.25: We have released the [QvQ-72B-Preview](https://huggingface.co/Qwen/QVQ-72B-Preview). QvQ-72B-Preview is an experimental research model, focusing on enhancing visual reasoning capabilities. For more details, please check our [blog](https://qwenlm.github.io/blog/qvq-72b-preview/)!
* 2024.09.19: The instruction-tuned [Qwen2-VL-72B model](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct) and its quantized version [[AWQ](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct-AWQ), [GPTQ-Int4](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int4), [GPTQ-Int8](https://huggingface.co/Qwen/Qwen2-VL-72B-Instruct-GPTQ-Int8)] are now available. We have also released the [Qwen2-VL paper](https://arxiv.org/pdf/2409.12191) simultaneously.
* 2024.08.30: We have released the [Qwen2-VL series](https://huggingface.co/collections/Qwen/qwen2-vl-66cee7455501d7126940800d). The 2B and 7B models are now available, and the 72B model for open source is coming soon. For more details, please check our [blog](https://qwenlm.github.io/blog/qwen2-vl/)!
## Performance
### Visual Tasks
<div style="display: flex; justify-content: center; gap: 16px; flex-wrap: wrap;">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_nothinking_vl.jpg" width="24%" />
<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-VL/table_thinking_vl_.jpg" width="24%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_nothinking_vl-30a3.jpg" width="26%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_thinking_vl_30A3.jpg" width="22.5%" />
</div>
<div style="display: flex; justify-content: center; gap: 16px; flex-wrap: wrap;">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_2b_32b_vl_instruct.jpg" width="30%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_2b_32b_vl_thinking.jpg" width="24%" />
</div>
### Text-Centric Tasks
<div style="display: flex; justify-content: center; gap: 16px; flex-wrap: wrap;">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_nothinking_text.jpg" width="30%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_thinking_text.jpg" width="32%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/table_nothinking_text-30a3.jpg" width="30%" />
</div>
<div style="display: flex; justify-content: center; gap: 16px; flex-wrap: wrap;">
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_4b_8b_text_instruct.jpg" width="33%" />
<img src="https://qianwen-res.oss-accelerate.aliyuncs.com/Qwen3-VL/qwen3vl_4b_8b_text_thinking.jpg" width="28%" />
</div>
## Cookbooks
We are preparing [cookbooks](https://github.com/QwenLM/Qwen3-VL/tree/main/cookbooks) for many capabilities, including recognition, localization, document parsing, video understanding, key information extraction, and more. Welcome to learn more!
| Cookbook | Description | Open |
| -------- | ----------- | ---- |
| [Omni Recognition](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/omni_recognition.ipynb) | Not only identify animals, plants, people, and scenic spots but also recognize various objects such as cars and merchandise. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/omni_recognition.ipynb) |
| [Powerful Document Parsing Capabilities](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/document_parsing.ipynb) | The parsing of documents has reached a higher level, including not only text but also layout position information and our Qwen HTML format. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/document_parsing.ipynb) |
| [Precise Object Grounding Across Formats](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/2d_grounding.ipynb) | Using relative position coordinates, it supports both boxes and points, allowing for diverse combinations of positioning and labeling tasks. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/2d_grounding.ipynb) |
| [General OCR and Key Information Extraction](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/ocr.ipynb) | Stronger text recognition capabilities in natural scenes and multiple languages, supporting diverse key information extraction needs. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/ocr.ipynb) |
| [Video Understanding](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/video_understanding.ipynb) | Better video OCR, long video understanding, and video grounding. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/video_understanding.ipynb) |
| [Mobile Agent](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/mobile_agent.ipynb) | Locate and think for mobile phone control. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/mobile_agent.ipynb) |
| [Computer-Use Agent](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/computer_use.ipynb) | Locate and think for controlling computers and Web. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/computer_use.ipynb) |
| [3D Grounding](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/3d_grounding.ipynb) | Provide accurate 3D bounding boxes for both indoor and outdoor objects. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/3d_grounding.ipynb) |
| [Thinking with Images](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/think_with_images.ipynb) | Utilize image_zoom_in_tool and search_tool to facilitate the model’s precise comprehension of fine-grained visual details within images. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/think_with_images.ipynb) |
| [MultiModal Coding](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/mmcode.ipynb) | Generate accurate code based on rigorous comprehension of multimodal information. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/mmcode.ipynb) |
| [Long Document Understanding](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/long_document_understanding.ipynb) | Achieve rigorous semantic comprehension of ultra-long documents. | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/long_document_understanding.ipynb) |
| [Spatial Understanding](https://github.com/QwenLM/Qwen3-VL/blob/main/cookbooks/spatial_understanding.ipynb) | See, understand and reason about the spatial information | [![Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenLM/Qwen3-VL/blob/main/cookbooks/spatial_understanding.ipynb) |
## Quickstart
Below, we provide simple examples to show how to use Qwen3-VL with 🤖 ModelScope and 🤗 Transformers.
```
# The Qwen3-VL model requires transformers >= 4.57.0
pip install "transformers>=4.57.0"
```
### 🤖 ModelScope
We strongly advise users especially those in mainland China to use ModelScope. `snapshot_download` can help you solve issues concerning downloading checkpoints.
### Using 🤗 Transformers to Chat
Here we show a code snippet to show you how to use the chat model with `transformers`:
```python
from transformers import AutoModelForImageTextToText, AutoProcessor
# default: Load the model on the available device(s)
model = AutoModelForImageTextToText.from_pretrained(
"Qwen/Qwen3-VL-235B-A22B-Instruct", dtype="auto", device_map="auto"
)
# We recommend enabling flash_attention_2 for better acceleration and memory saving, especially in multi-image and video scenarios.
# model = AutoModelForImageTextToText.from_pretrained(
# "Qwen/Qwen3-VL-235B-A22B-Instruct",
# dtype=torch.bfloat16,
# attn_implementation="flash_attention_2",
# device_map="auto",
# )
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-235B-A22B-Instruct")
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
},
{"type": "text", "text": "Describe this image."},
],
}
]
# Preparation for inference
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt"
)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
<!-- <details>
<summary>Minimum VRAM requirements</summary>
| Precision | Qwen2.5-VL-3B | Qwen2.5-VL-7B | Qwen2.5-VL-72B |
|-----------|------------| --------- | -------- |
| FP32 | 11.5 GB | 26.34 GB | 266.21 GB |
| BF16 | 5.75 GB | 13.17 GB | 133.11 GB |
| INT8 | 2.87 GB | 6.59 GB | 66.5 GB |
| INT4 | 1.44 GB | 3.29 GB | 33.28 GB |
Note: The table above presents the theoretical minimum video memory requirements for inference with `transformers`; however, in practice, the actual memory usage is typically at least 1.2 times higher. For more information, see the linked resource [here](https://huggingface.co/docs/accelerate/main/en/usage_guides/model_size_estimator).
</details> -->
<details>
<summary>Multi image inference</summary>
```python
# Messages containing multiple images and a text query
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": "file:///path/to/image1.jpg"},
{"type": "image", "image": "file:///path/to/image2.jpg"},
{"type": "text", "text": "Identify the similarities between these images."},
],
}
]
# Preparation for inference
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt"
)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
<details>
<summary>Video inference</summary>
```python
# Messages containing a video url(or a local path) and a text query
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4",
},
{"type": "text", "text": "Describe this video."},
],
}
]
# Preparation for inference
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt"
)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
<details>
<summary>Batch inference</summary>
```python
# for batch generation, padding_side should be set to left!
processor.tokenizer.padding_side = 'left'
# Sample messages for batch inference
messages1 = [
{
"role": "user",
"content": [
{"type": "image", "image": "file:///path/to/image1.jpg"},
{"type": "image", "image": "file:///path/to/image2.jpg"},
{"type": "text", "text": "What are the common elements in these pictures?"},
],
}
]
messages2 = [
{"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
{"role": "user", "content": [{"type": "text", "text": "Who are you?"}]},
]
# Combine messages for batch processing
messages = [messages1, messages2]
# Preparation for inference
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
padding=True # padding should be set for batch generation!
)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
<details>
<summary>Pixel Control via Official Processor</summary>
Using the official HF processor, we can conveniently control the budget of visual tokens. Since the Qwen3-VL processor separates image and video processing, we can independently configure the pixel budget for each modality.
- **For the image processor**:
The parameter `size['longest_edge']` originally corresponds to `max_pixels`, which defines the maximum number of pixels allowed for an image (i.e., for an image of height H and width W, H × W must not exceed `max_pixels`; image channels are ignored for simplicity).
Similarly, `size['shortest_edge']` corresponds to `min_pixels`, specifying the minimum allowable pixel count for an image.
- **For the video processor**:
The interpretation differs slightly. `size['longest_edge']` represents the maximum total number of pixels across all frames in a video — for a video of shape T×H×W, the product T×H×W must not exceed `size['longest_edge']`.
Similarly, `size['shortest_edge']` sets the minimum total pixel budget for the video.
```python
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-235B-A22B-Instruct")
# budget for image processor, since the compression ratio is 32 for Qwen3-VL, we can set the number of visual tokens of a single image to 256-1280
processor.image_processor.size = {"longest_edge": 1280*32*32, "shortest_edge": 256*32*32}
# budget for video processor, we can set the number of visual tokens of a single video to 256-16384
processor.video_processor.size = {"longest_edge": 16384*32*32, "shortest_edge": 256*32*32}
```
- You can further control the **sample fps** or **sample frames** of video, as shown below.
```python
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4",
},
{"type": "text", "text": "Describe this video."},
],
}
]
# for video input, we can further control the fps or num_frames. \
# defaultly, fps is set to 2
# set fps = 4
inputs = processor.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
return_tensors="pt",
fps=4
)
inputs = inputs.to(model.device)
# set num_frames = 128 and overwrite the fps to None!
# inputs = processor.apply_chat_template(
# messages,
# tokenize=True,
# add_generation_prompt=True,
# return_dict=True,
# return_tensors="pt",
# num_frames=128,
# fps=None,
# )
# inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
### New `qwen-vl-utils` Usage
With the latest `qwen-vl-utils` toolkit (backward compatible with Qwen2.5-VL), you can control pixel constraints per visual input.
```bash
pip install qwen-vl-utils==0.0.14
# It's highly recommended to use `[decord]` feature for faster video loading.
# pip install qwen-vl-utils[decord]
```
Compared to previous version, the new `qwen-vl-utils` introduces:
- "image_patch_size": `14` for Qwen2.5-VL and `16` for Qwen3-VL. Default set to `14`.
- "return_video_metadata"(Qwen3-VL only): Due to the new video processor, if True, each video returns as (video_tensor, video_metadata). Default set to `False`.
```python
# for Qwen2.5VL, you can simply call
images, videos, video_kwargs = process_vision_info(messages, return_video_kwargs=True)
# For Qwen3VL series, you should call
images, videos, video_kwargs = process_vision_info(messages, image_patch_size=16, return_video_kwargs=True, return_video_metadata=True)
```
📌 Note: Since `qwen-vl-utils` already resizes images/videos, pass `do_resize=False` to the processor to avoid duplicate resizing.
<details>
<summary>Process Images</summary>
For input images, we support local files, base64, and URLs.
```python
# You can directly insert a local file path, a URL, or a base64-encoded image into the position where you want in the text.
## Local file path
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": "file:///path/to/your/image.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
## Image URL
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": "http://path/to/your/image.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
## Base64 encoded image
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": "data:image;base64,/9j/..."},
{"type": "text", "text": "Describe this image."},
],
}
]
```
We provide two methods for fine-grained control over the image size input to the model:
- Specify exact dimensions: Directly set resized_height and resized_width. These values will be rounded to the nearest multiple of 32 (32 for Qwen3VL, 28 for Qwen2.5VL).
- Define min_pixels and max_pixels: Images will be resized to maintain their aspect ratio within the range of min_pixels and max_pixels
```python
from transformers import AutoModelForImageTextToText, AutoProcessor
from qwen_vl_utils import process_vision_info
model = AutoModelForImageTextToText.from_pretrained(
"Qwen/Qwen3-VL-235B-A22B-Instruct", dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-235B-A22B-Instruct")
# resized_height and resized_width
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
"resized_height": 280,
"resized_width": 420,
},
{"type": "text", "text": "Describe this image."},
],
}
]
# min_pixels and max_pixels
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
"min_pixels": 50176,
"max_pixels": 50176,
},
{"type": "text", "text": "Describe this image."},
],
}
]
# Preparation for inference with qwen-vl-utils
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
images, videos = process_vision_info(messages, image_patch_size=16)
# since qwen-vl-utils has resize the images/videos, \
# we should pass do_resize=False to avoid duplicate operation in processor!
inputs = processor(text=text, images=images, videos=videos, do_resize=False, return_tensors="pt")
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
<details>
<summary>Process Videos</summary>
For input videos, we support images lists, local path and url.
```python
# Messages containing a images list as a video and a text query
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": [
"file:///path/to/frame1.jpg",
"file:///path/to/frame2.jpg",
"file:///path/to/frame3.jpg",
"file:///path/to/frame4.jpg",
],
'sample_fps':'1', # sample_fps: frame sampling rate (frames per second), used to determine timestamps for each frame
},
{"type": "text", "text": "Describe this video."},
],
}
]
# Messages containing a local video path and a text query
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": "file:///path/to/video1.mp4",
"max_pixels": 360 * 420,
"fps": 1.0,
},
{"type": "text", "text": "Describe this video."},
],
}
]
# Messages containing a video url and a text query
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4",
"min_pixels": 4 * 32 * 32,
"max_pixels": 256 * 32 * 32,
"total_pixels": 20480 * 32 * 32,
},
{"type": "text", "text": "Describe this video."},
],
}
]
```
We recommend setting appropriate values for the `min_pixels` and `max_pixels` parameters based on available GPU memory and the specific application scenario to restrict the resolution of individual frames in the video.
Alternatively, you can use the `total_pixels` parameter to limit the total number of tokens in the video (it is recommended to set this value below 24576 * 32 * 32 to avoid excessively long input sequences). For more details on parameter usage and processing logic, please refer to the `fetch_video` function in `qwen_vl_utils/vision_process.py`.
```python
from transformers import AutoModelForImageTextToText, AutoProcessor
from qwen_vl_utils import process_vision_info
model = AutoModelForImageTextToText.from_pretrained(
"Qwen/Qwen3-VL-235B-A22B-Instruct", dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-235B-A22B-Instruct")
messages = [
{
"role": "user",
"content": [
{
"type": "video",
"video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4",
"min_pixels": 4 * 32 * 32,
"max_pixels": 256 * 32 * 32,
"total_pixels": 20480 * 32 * 32,
},
{"type": "text", "text": "Describe this video."},
],
}
]
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
images, videos, video_kwargs = process_vision_info(messages, image_patch_size=16, return_video_kwargs=True, return_video_metadata=True)
# split the videos and according metadatas
if videos is not None:
videos, video_metadatas = zip(*videos)
videos, video_metadatas = list(videos), list(video_metadatas)
else:
video_metadatas = None
# since qwen-vl-utils has resize the images/videos, \
# we should pass do_resize=False to avoid duplicate operation in processor!
inputs = processor(text=text, images=images, videos=videos, video_metadata=video_metadatas, return_tensors="pt", do_resize=False, **video_kwargs)
inputs = inputs.to(model.device)
# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128)
generated_ids_trimmed = [
out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]
output_text = processor.batch_decode(
generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
)
print(output_text)
```
</details>
<details>
<summary>Video Backends and URL Compatibility</summary>
Currently, `qwen-vl-utils` supports three video decoding backends: `torchvision`, `decord`, and `torchcodec`. While `decord` and `torchcodec` generally offer significantly faster decoding speeds compared to `torchvision`, we recommend using `torchcodec`. This is because `decord` has known issues, such as decoding hangs, and its project is no longer actively maintained.
- For `decord`, if you are not using Linux, you might not be able to install `decord` from PyPI. In that case, you can use `pip install qwen-vl-utils` which will fall back to using torchvision for video processing. However, you can still [install decord from source](https://github.com/dmlc/decord?tab=readme-ov-file#install-from-source) to get decord used when loading video.
- To use `torchcodec` as the backend for video decoding, follow the installation instructions provided in the official [torchcodec repository](https://github.com/pytorch/torchcodec/tree/main?tab=readme-ov-file#installing-torchcodec) and install it manually. Note that `torchcodec` depends on FFmpeg for decoding functionality.
Video URL compatibility is primarily determined by the version of the third-party library being used. For more details, refer to the table below. If you prefer not to use the default backend, you can switch it by setting `FORCE_QWENVL_VIDEO_READER` to `torchvision`, `decord`, or `torchcodec`.
| Backend | HTTP | HTTPS |
|-------------|------|-------|
| torchvision >= 0.19.0 | ✅ | ✅ |
| torchvision < 0.19.0 | ❌ | ❌ |
| decord | ✅ | ❌ |
| torchcodec | ✅ | ✅ |
</details>
### More Usage Tips
#### Add ids for Multiple Visual Inputs
By default, images and video content are directly included in the conversation. When handling multiple images, it's helpful to add labels to the images and videos for better reference. Users can control this behavior with the following settings:
<details>
<summary>Add vision ids</summary>
```python
conversation = [
{
"role": "user",
"content": [{"type": "image"}, {"type": "text", "text": "Hello, how are you?"}],
},
{
"role": "assistant",
"content": "I'm doing well, thank you for asking. How can I assist you today?",
},
{
"role": "user",
"content": [
{"type": "text", "text": "Can you describe these images and video?"},
{"type": "image"},
{"type": "image"},
{"type": "video"},
{"type": "text", "text": "These are from my vacation."},
],
},
{
"role": "assistant",
"content": "I'd be happy to describe the images and video for you. Could you please provide more context about your vacation?",
},
{
"role": "user",
"content": "It was a trip to the mountains. Can you see the details in the images and video?",
},
]
# default:
prompt_without_id = processor.apply_chat_template(
conversation, add_generation_prompt=True
)
# Excepted output: '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n<|im_start|>user\n<|vision_start|><|image_pad|><|vision_end|>Hello, how are you?<|im_end|>\n<|im_start|>assistant\nI'm doing well, thank you for asking. How can I assist you today?<|im_end|>\n<|im_start|>user\nCan you describe these images and video?<|vision_start|><|image_pad|><|vision_end|><|vision_start|><|image_pad|><|vision_end|><|vision_start|><|video_pad|><|vision_end|>These are from my vacation.<|im_end|>\n<|im_start|>assistant\nI'd be happy to describe the images and video for you. Could you please provide more context about your vacation?<|im_end|>\n<|im_start|>user\nIt was a trip to the mountains. Can you see the details in the images and video?<|im_end|>\n<|im_start|>assistant\n'
# add ids
prompt_with_id = processor.apply_chat_template(
conversation, add_generation_prompt=True, add_vision_id=True
)
# Excepted output: '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n<|im_start|>user\nPicture 1: <|vision_start|><|image_pad|><|vision_end|>Hello, how are you?<|im_end|>\n<|im_start|>assistant\nI'm doing well, thank you for asking. How can I assist you today?<|im_end|>\n<|im_start|>user\nCan you describe these images and video?Picture 2: <|vision_start|><|image_pad|><|vision_end|>Picture 3: <|vision_start|><|image_pad|><|vision_end|>Video 1: <|vision_start|><|video_pad|><|vision_end|>These are from my vacation.<|im_end|>\n<|im_start|>assistant\nI'd be happy to describe the images and video for you. Could you please provide more context about your vacation?<|im_end|>\n<|im_start|>user\nIt was a trip to the mountains. Can you see the details in the images and video?<|im_end|>\n<|im_start|>assistant\n'
```
</details>
#### Flash-Attention 2 to speed up generation
First, make sure to install the latest version of Flash Attention 2:
```bash
pip install -U flash-attn --no-build-isolation
```
Also, you should have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of the [flash attention repository](https://github.com/Dao-AILab/flash-attention). FlashAttention-2 can only be used when a model is loaded in `torch.float16` or `torch.bfloat16`.
To load and run a model using Flash Attention-2, simply add `attn_implementation="flash_attention_2"` when loading the model as follows:
```python
import torch
from transformers import AutoModelForImageTextToText
model = AutoModelForImageTextToText.from_pretrained(
"Qwen/Qwen3-VL-235B-A22B-Instruct",
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
)
```
#### Processing Long Texts
The current `config.json` is set for context length up to 256K tokens.
To handle extensive inputs exceeding 256K tokens, we utilize [YaRN](https://arxiv.org/abs/2309.00071), a technique for enhancing model length extrapolation, ensuring optimal performance on lengthy texts.
For supported frameworks (currently transformers and vLLM), you could modify `max_position_embeddings` and `rope_scaling` in `config.json` to enable YaRN:
```
{
"max_position_embeddings": 1000000,
...,
"rope_scaling": {
"rope_type": "yarn",
"mrope_section": [
24,
20,
20
],
"mrope_interleaved": true,
"factor": 3.0,
"original_max_position_embeddings": 262144
},
...
}
```
When using vLLM for serving, you can also enable YaRN by adding the additional arguments `--rope-scaling` and `--max-model-len`.
```
vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct --rope-scaling '{"rope_type":"yarn","factor":3.0,"original_max_position_embeddings": 262144,"mrope_section":[24,20,20],"mrope_interleaved": true}' --max-model-len 1000000
```
> Because Interleaved-MRoPE’s position IDs grow more slowly than vanilla RoPE, use a **smaller scaling factor**. For example, to support 1M context with 256K context length, set factor=2 or 3 — not 4.
### Try Qwen3-VL-235B-A22 with API!
To explore Qwen3-VL-235B-A22, a more fascinating multimodal model, we encourage you to test our cutting-edge API service. Let's start the exciting journey right now!
```python
from openai import OpenAI
# set your DASHSCOPE_API_KEY here
DASHSCOPE_API_KEY = ""
client = OpenAI(
api_key=DASHSCOPE_API_KEY,
base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
)
completion = client.chat.completions.create(
model="qwen3-vl-235b-a22b-instruct",
messages=[{"role": "user", "content": [
{"type": "image_url",
"image_url": {"url": "https://dashscope.oss-cn-beijing.aliyuncs.com/images/dog_and_girl.jpeg"}},
{"type": "text", "text": "这是什么"},
]}]
)
print(completion.model_dump_json())
```
For more usage, please refer to the tutorial at [aliyun](https://help.aliyun.com/zh/model-studio/developer-reference/qwen-vl-api).
### Web UI Example
In this section, we provide instructions for users to build a web-based user interface (UI) demo. This UI demo allows users to interact with a predefined model or application through a web browser. Follow the steps below to get started.
Install the required dependencies by running the following command:
```bash
pip install -r requirements_web_demo.txt
```
Launch a browser-based UI to interact with the model:
```bash
python web_demo_mm.py -c /your/path/to/qwen3vl/weight
```
After running the command, you’ll see a link generated in the terminal similar to this:
```
Running on local: http://127.0.0.1:7860/
```
Open the link in your browser to interact with the model — try text, images, or other features. For a quick start, you can also use our pre-built Docker image:
```
cd docker && bash run_web_demo.sh -c /your/path/to/qwen3vl/weight --port 8881
```
## Deployment
We recommend using vLLM for fast Qwen3-VL deployment and inference. You need to install `vllm>=0.11.0` to enable Qwen3-VL support. You can also use our [official docker image](#-docker).
Please check [vLLM official documentation](https://docs.vllm.ai/en/latest/serving/multimodal_inputs.html) for more details about online serving and offline inference for multimodal models.
### Installation
```bash
pip install accelerate
pip install qwen-vl-utils==0.0.14
# Install the latest version of vLLM 'vllm>=0.11.0'
uv pip install -U vllm
```
### Online Serving
You can start either a vLLM or SGLang server to serve LLMs efficiently, and then access it using an OpenAI-style API.
The following launch command is applicable to H100/H200; for more efficient deployment or deployment on other GPUs, please refer to the [vLLM community guide](https://docs.vllm.ai/projects/recipes/en/latest/Qwen/Qwen3-VL.html).
* vLLM server
```shell
# Efficient inference with FP8 checkpoint
# Requires NVIDIA H100+ and CUDA 12+
vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct-FP8 \
--tensor-parallel-size 8 \
--mm-encoder-tp-mode data \
--enable-expert-parallel \
--async-scheduling \
--media-io-kwargs '{"video": {"num_frames": -1}}' \
--host 0.0.0.0 \
--port 22002
```
* SGLang server
```
python -m sglang.launch_server \
--model-path Qwen/Qwen3-VL-235B-A22B-Instruct \
--host 0.0.0.0 \
--port 22002 \
--tp 4
```
* Image Request Example
```python
import time
from openai import OpenAI
client = OpenAI(
api_key="EMPTY",
base_url="http://127.0.0.1:22002/v1",
timeout=3600
)
messages = [
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": "https://ofasys-multimodal-wlcb-3-toshanghai.oss-accelerate.aliyuncs.com/wpf272043/keepme/image/receipt.png"
}
},
{
"type": "text",
"text": "Read all the text in the image."
}
]
}
]
start = time.time()
response = client.chat.completions.create(
model="Qwen/Qwen3-VL-235B-A22B-Instruct-FP8",
messages=messages,
max_tokens=2048
)
print(f"Response costs: {time.time() - start:.2f}s")
print(f"Generated text: {response.choices[0].message.content}")
```
* Video Request Example
```python
import time
from openai import OpenAI
client = OpenAI(
api_key="EMPTY",
base_url="http://127.0.0.1:22002/v1",
timeout=3600
)
messages = [
{
"role": "user",
"content": [
{
"type": "video_url",
"video_url": {
"url": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4"
}
},
{
"type": "text",
"text": "How long is this video?"
}
]
}
]
start = time.time()
# When vLLM is launched with `--media-io-kwargs '{"video": {"num_frames": -1}}'`,
# video frame sampling can be configured via `extra_body` (e.g., by setting `fps`).
# This feature is currently supported only in vLLM.
#
# By default, `fps=2` and `do_sample_frames=True`.
# With `do_sample_frames=True`, you can customize the `fps` value to set your desired video sampling rate.
response = client.chat.completions.create(
model="Qwen/Qwen3-VL-235B-A22B-Instruct-FP8",
messages=messages,
max_tokens=2048,
extra_body={"mm_processor_kwargs": {"fps": 2, "do_sample_frames": True}}
)
print(f"Response costs: {time.time() - start:.2f}s")
print(f"Generated text: {response.choices[0].message.content}")
```
### Offline Inference
You can also use vLLM or SGLang to inference Qwen3-VL locally:
* vLLM Examples
``` python
# -*- coding: utf-8 -*-
import torch
from qwen_vl_utils import process_vision_info
from transformers import AutoProcessor
from vllm import LLM, SamplingParams
import os
os.environ['VLLM_WORKER_MULTIPROC_METHOD'] = 'spawn'
def prepare_inputs_for_vllm(messages, processor):
text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
# qwen_vl_utils 0.0.14+ reqired
image_inputs, video_inputs, video_kwargs = process_vision_info(
messages,
image_patch_size=processor.image_processor.patch_size,
return_video_kwargs=True,
return_video_metadata=True
)
print(f"video_kwargs: {video_kwargs}")
mm_data = {}
if image_inputs is not None:
mm_data['image'] = image_inputs
if video_inputs is not None:
mm_data['video'] = video_inputs
return {
'prompt': text,
'multi_modal_data': mm_data,
'mm_processor_kwargs': video_kwargs
}
if __name__ == '__main__':
# messages = [
# {
# "role": "user",
# "content": [
# {
# "type": "video",
# "video": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen2-VL/space_woaudio.mp4",
# },
# {"type": "text", "text": "这段视频有多长"},
# ],
# }
# ]
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://ofasys-multimodal-wlcb-3-toshanghai.oss-accelerate.aliyuncs.com/wpf272043/keepme/image/receipt.png",
},
{"type": "text", "text": "Read all the text in the image."},
],
}
]
# TODO: change to your own checkpoint path
checkpoint_path = "Qwen/Qwen3-VL-235B-A22B-Instruct-FP8"
processor = AutoProcessor.from_pretrained(checkpoint_path)
inputs = [prepare_inputs_for_vllm(message, processor) for message in [messages]]
llm = LLM(
model=checkpoint_path,
mm_encoder_tp_mode="data",
enable_expert_parallel=True,
tensor_parallel_size=torch.cuda.device_count(),
seed=0
)
sampling_params = SamplingParams(
temperature=0,
max_tokens=1024,
top_k=-1,
stop_token_ids=[],
)
for i, input_ in enumerate(inputs):
print()
print('=' * 40)
print(f"Inputs[{i}]: {input_['prompt']=!r}")
print('\n' + '>' * 40)
outputs = llm.generate(inputs, sampling_params=sampling_params)
for i, output in enumerate(outputs):
generated_text = output.outputs[0].text
print()
print('=' * 40)
print(f"Generated text: {generated_text!r}")
```
* SGLang Examples
```python
import time
from PIL import Image
from sglang import Engine
from qwen_vl_utils import process_vision_info
from transformers import AutoProcessor, AutoConfig
if __name__ == "__main__":
# TODO: change to your own checkpoint path
checkpoint_path = "Qwen/Qwen3-VL-235B-A22B-Instruct"
processor = AutoProcessor.from_pretrained(checkpoint_path)
messages = [
{
"role": "user",
"content": [
{
"type": "image",
"image": "https://ofasys-multimodal-wlcb-3-toshanghai.oss-accelerate.aliyuncs.com/wpf272043/keepme/image/receipt.png",
},
{"type": "text", "text": "Read all the text in the image."},
],
}
]
text = processor.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
image_inputs, _ = process_vision_info(messages, image_patch_size=processor.image_processor.patch_size)
llm = Engine(
model_path=checkpoint_path,
enable_multimodal=True,
mem_fraction_static=0.8,
tp_size=4,
attention_backend="fa3",
context_length=10240,
disable_cuda_graph=True,
)
start = time.time()
sampling_params = {"max_new_tokens": 1024}
response = llm.generate(prompt=text, image_data=image_inputs, sampling_params=sampling_params)
print(f"Response costs: {time.time() - start:.2f}s")
print(f"Generated text: {response['text']}")
```
## Evaluation Reproduction
To facilitate faithful reproduction of our reported results, we summarize our official evaluation settings below.
- Inference runtime: [vLLM](https://github.com/vllm-project/vllm)
- Evaluation frameworks: [VLMEvalKit](https://github.com/open-compass/VLMEvalKit), [lmms-eval](https://github.com/EvolvingLMMs-Lab/lmms-eval)
- Notes:
- For a few benchmarks, we slightly modified the evaluation prompts; detailed changes will be documented in the upcoming technical report.
- A small number of benchmarks are internally constructed; we plan to release the code and reproduction assets afterwards.
### Generation Hyperparameters
#### Instruct models
```bash
export greedy='false'
export seed=3407
export top_p=0.8
export top_k=20
export temperature=0.7
export repetition_penalty=1.0
export presence_penalty=1.5
export out_seq_length=32768
```
#### Thinking models
```bash
export greedy='false'
export seed=1234
export top_p=0.95
export top_k=20
export repetition_penalty=1.0
export presence_penalty=0.0
export temperature=0.6
export out_seq_length=40960
```
## 🐳 Docker
To simplify the deploy process, we provide docker images with pre-build environments: [qwenllm/qwenvl](https://hub.docker.com/r/qwenllm/qwenvl). You only need to install the driver and download model files to launch demos.
```bash
docker run --gpus all --ipc=host --network=host --rm --name qwen3vl -it qwenllm/qwenvl:qwen3vl-cu128 bash
```
## Citation
If you find our paper and code useful in your research, please consider giving a star :star: and citation :pencil: :)
```BibTeX
@article{Qwen3-VL,
title={Qwen3-VL Technical Report},
author={Shuai Bai and Yuxuan Cai and Ruizhe Chen and Keqin Chen and Xionghui Chen and Zesen Cheng and Lianghao Deng and Wei Ding and Chang Gao and Chunjiang Ge and Wenbin Ge and Zhifang Guo and Qidong Huang and Jie Huang and Fei Huang and Binyuan Hui and Shutong Jiang and Zhaohai Li and Mingsheng Li and Mei Li and Kaixin Li and Zicheng Lin and Junyang Lin and Xuejing Liu and Jiawei Liu and Chenglong Liu and Yang Liu and Dayiheng Liu and Shixuan Liu and Dunjie Lu and Ruilin Luo and Chenxu Lv and Rui Men and Lingchen Meng and Xuancheng Ren and Xingzhang Ren and Sibo Song and Yuchong Sun and Jun Tang and Jianhong Tu and Jianqiang Wan and Peng Wang and Pengfei Wang and Qiuyue Wang and Yuxuan Wang and Tianbao Xie and Yiheng Xu and Haiyang Xu and Jin Xu and Zhibo Yang and Mingkun Yang and Jianxin Yang and An Yang and Bowen Yu and Fei Zhang and Hang Zhang and Xi Zhang and Bo Zheng and Humen Zhong and Jingren Zhou and Fan Zhou and Jing Zhou and Yuanzhi Zhu and Ke Zhu},
journal={arXiv preprint arXiv:2511.21631},
year={2025}
}
@article{Qwen2.5-VL,
title={Qwen2.5-VL Technical Report},
author={Bai, Shuai and Chen, Keqin and Liu, Xuejing and Wang, Jialin and Ge, Wenbin and Song, Sibo and Dang, Kai and Wang, Peng and Wang, Shijie and Tang, Jun and Zhong, Humen and Zhu, Yuanzhi and Yang, Mingkun and Li, Zhaohai and Wan, Jianqiang and Wang, Pengfei and Ding, Wei and Fu, Zheren and Xu, Yiheng and Ye, Jiabo and Zhang, Xi and Xie, Tianbao and Cheng, Zesen and Zhang, Hang and Yang, Zhibo and Xu, Haiyang and Lin, Junyang},
journal={arXiv preprint arXiv:2502.13923},
year={2025}
}
@article{Qwen2-VL,
title={Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution},
author={Wang, Peng and Bai, Shuai and Tan, Sinan and Wang, Shijie and Fan, Zhihao and Bai, Jinze and Chen, Keqin and Liu, Xuejing and Wang, Jialin and Ge, Wenbin and Fan, Yang and Dang, Kai and Du, Mengfei and Ren, Xuancheng and Men, Rui and Liu, Dayiheng and Zhou, Chang and Zhou, Jingren and Lin, Junyang},
journal={arXiv preprint arXiv:2409.12191},
year={2024}
}
@article{Qwen-VL,
title={Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond},
author={Bai, Jinze and Bai, Shuai and Yang, Shusheng and Wang, Shijie and Tan, Sinan and Wang, Peng and Lin, Junyang and Zhou, Chang and Zhou, Jingren},
journal={arXiv preprint arXiv:2308.12966},
year={2023}
}
```
<br>
cookbooks/2d_grounding.ipynb 0 → 100644
View file @ 3d735feb
This source diff could not be displayed because it is too large. You can view the blob instead.
cookbooks/3d_grounding.ipynb 0 → 100644
View file @ 3d735feb
This source diff could not be displayed because it is too large. You can view the blob instead.
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment