🎉 FastDeploy 2.3 多模态模型部署性能极致优化，实测挑战赛全面升级！

[ethan7zhanghx]

FastDeploy2.3版本现已正式上线，新增ERNIE-4.5-VL-Thinking、PaddleOCR-VL模型的支持！
FastDeploy 2.3 is officially released, now with native support for ERNIE-4.5-VL-Thinking and PaddleOCR-VL models!

完成以下任一实测挑战，即可获得最高1万奖金(总奖池10万元)及APPLE 代金卡、FastDeploy骨瓷马克杯等官方周边！
Complete any of the challenges below for a chance to win up to ¥10,000 (from a total ¥100,000 prize pool), plus Apple gift cards and official FastDeploy merch!

赛题说明 Challenge Overview

🎯 赛道一：FastDeploy2.3 x 多模态模型 推理性能实测

Track 1: FastDeploy 2.3 × Multimodal Model Inference Performance Test

🧑‍💻 任务描述 Task Description

基于FastDeploy2.3 在任一算力环境成功部署PaddleOCR-VL或ERNIE-4.5-VL-Thinking，并完成3次基于不同真实多模数据的高效推理，提供实测截图（需含结果数据、性能数据等内容）。
Deploy either PaddleOCR-VL or ERNIE-4.5-VL-Thinking using FastDeploy 2.3 on any computing environment, and complete 3 high-efficiency inference runs with different real multimodal data.
Submit screenshots that include inference results and performance metrics as proof of completion.

🎯 赛道二：FastDeploy2.3 x 文本生成大模型 推理性能实测

🎯 Track 2: FastDeploy 2.3 × Large Language Model (LLM) Inference Performance Test

🧑‍💻 任务描述 Task Description
基于FastDeploy2.3 在任一算力环境完成ERNIE 4.5系列、DeepSeek系列、Qwen系列等主流大模型或其衍生模型的部署，并成功运行3个基于真实数据的推理任务，提供实测截图（需含结果数据、性能数据等内容）。
Use FastDeploy 2.3 to deploy any mainstream large language model — such as the ERNIE 4.5, DeepSeek, or Qwen series — or their derived variants, on any computing environment. Successfully complete three inference tasks with real-world data and provide screenshots including inference results and performance metrics.

该赛题不限制模型尺寸，奖金随部署难度递增，如完成千亿级大模型高性能部署则奖金加倍！
This track has no restriction on model size, and the reward increases with deployment difficulty.
Successfully deploying a trillion-parameter LLM with high performance will double your prize!

🎯 赛道三：FastDeploy2.3 x 新硬件专题挑战

🎯 Track 3: FastDeploy 2.3 × Emerging Hardware Challenge

🧑‍💻 任务描述 Task Description
基于FastDeploy2.3 在昆仑芯XPU、海光DCU等新硬件平台完成任一大模型推理部署服务，并成功运行3个基于真实数据的推理任务，提供实测截图（需含结果数据、性能数据等内容）。
Use FastDeploy 2.3 to deploy any large model inference service on new-generation hardware platforms such as Baidu Kunlun XPU or Hygon DCU. Successfully run three inference tasks with real-world data and provide screenshots containing results and performance metrics.

🎯 赛道四：创意应用方向

🎯 Track 4: Creative Application Development

🧑‍💻 任务描述 Task Description
与任一实际应用场景结合（如工业、医疗、交通、游戏、传媒等），将此前部署好的模型服务集成为有具体功能的应用，并使用真实数据完成3次推理验证
Integrate your previously deployed model services into a functional application aligned with any real-world scenario — such as industrial, medical, transportation, gaming, or media use cases. Run three inference validations with real data to demonstrate functionality and performance.

🎯赛道五：企业集成方向

🎯 Track 5: Enterprise Integration Track

🧑‍💻 任务描述
针对企业具体生产场景，将FD能力应用在具体场景解决方案中。该赛道侧重技术方案可行性评估，无需实际完成工业级部署。技术方案中需包含需求分析、模型及推理框架选型、场景硬件及数据条件等要素，奖金与星河产业应用创新奖不互斥。
Apply FastDeploy’s capabilities to a real-world enterprise production scenario, developing a technical solution for that context. This track focuses on feasibility evaluation of your proposed solution — industrial-level deployment is not required.
Your technical proposal should include requirement analysis, model and inference framework selection, hardware and data setup, and other relevant factors.

---

累计完成三项不同赛道的实测任务，即可获得额外奖金加成500元！
Complete tasks from three different tracks to earn an additional ¥500 bonus!

🚪 结果提交 Result Submission：

传送门： https://www.wjx.top/vm/rH2OyxL.aspx#
注意将您所有的作业文件（py文件、日志、截图、blog地址等）上传至问卷最后一题
Please upload all your artifacts (e.g., .py files, logs, screenshots, blog URLs, etc.) to the final question of the survey.

If you cannot access the survey link, you may email your submission to zhanghaoxin0819@gmail.com.

活动时间 Event Duration： 2025.11.05——2025.12.22

📕 参考教程 Reference Guides：

硬件及模型支持列表 Supported Hardware and Model List https://paddlepaddle.github.io/FastDeploy/zh/
官方安装文档 Installation Guide https://paddlepaddle.github.io/FastDeploy/zh/get_started/installation/

安装好FastDeploy 2.3后，用如下命令即可一键启动服务
After installing FastDeploy 2.3, you can start the service with a single command:

python -m fastdeploy.entrypoints.openai.api_server --model PaddlePaddle/PaddleOCR-VL --port 8188

服务启动后，您可以用下面的测试 demo 确认服务是否正常。
Once the service is running, you can use the following test demo to verify that it works properly.

from openai import OpenAI

client = OpenAI(
    api_key="EMPTY",
    base_url="http://0.0.0.0:8188/v1",
)
chat_response = client.chat.completions.create(
    model=client.models.list().data[0].id,
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": "https://paddle-model-ecology.bj.bcebos.com/paddlex/imgs/demo_image/ocr_demo.jpg"}},
            {"type": "text", "text": "OCR:"},
        ],
    }],
    temperature=1.0,
    max_tokens=2048,
    top_p=0,
)
print("结果:", chat_response.choices[0].message.content)

若您想在其他硬件上部署，或者希望深入了解最佳的部署方法，可参考我们的 PaddleOCR-VL部署文档
If you’d like to deploy on other hardware platforms or explore the optimal deployment methods,
please refer to our PaddleOCR-VL Deployment Guide:

https://paddlepaddle.github.io/FastDeploy/zh/best_practices/PaddleOCR-VL-0.9B/