Meet ERNIE 5, ERNIE 4.5 Open-source Series, and PaddleOCR here!
This challenge invites developers to innovate with ERNIE & PaddlePaddle's cutting-edge capabilities, pushing the limits of model customization and real-world applications.
LLaMA-Factory, Unsloth, Novita AI, CAMEL-AI, D-Robotics have partnered with Baidu to bring you this challenge. Your mission is to leverage the unique capabilities of ERNIE & PaddlePaddle to build intelligent and creative solutions. build a practical robot or creative application, fine-tune the model effectively to make it unique, or push the limits of model inference and deployment in the most imaginative and impactful ways.
Get Started
- ERNIE 5
-
ERNIE 4.5 Open-Source Seires
- Learn about the models: https://yiyan.baidu.com/blog/posts/ernie4.5/
- Huggingface: https://huggingface.co/collections/baidu/ernie-45
-
PaddleOCR
- Learn about the models: https://aistudio.baidu.com/paddleocr
- Huggingface: https://huggingface.co/PaddlePaddle/PaddleOCR-VL
Requirements
WHAT TO BUILD
We offer a range of tasks from beginner-friendly warm-ups to model and application development tasks. Just submit your project under one or more of the following tasks.
Warm-up Task
- Web Builder: Build a Web Page with PaddleOCR & ERNIE
- For all participants who use PaddleOCR-VL to extract text and layout from a PDF, convert the content into Markdown, then use ERNIE model to generate a web page, and finally deploy it on GitHub Pages.
Model-Building Tasks
- Best ERNIE Fine-Tune using Unsloth:
- For the best fine-tuned ERNIE model created using Unsloth, optimized for a specific, impactful task.
- Best ERNIE Fine-Tune using LLaMA-Factory:
- For the best fine-tuned ERNIE model created using LLaMA-Factory, optimized for a specific, impactful task.
- Best PaddleOCR-VL Fine-Tune:
- For the best fine-tuned PaddleOCR-VL model created, optimized for a specific, impactful task.
Application-Building Tasks
Build an innovative application powered by ERNIE/PaddleOCR and their ecosystem partners, choose your application theme and technology stack from the following directions:
- Best ERNIE Multimodal Application
- For the best ERNIE Multimodal Application using Baidu AI Studio API
- Best ERNIE Multimodal Application
- For the best ERNIE Multimodal Application using Novita API
- Best in Edge AI & Robotics
- For the best Edge AI & Robotics Applications integrating ERNIE/PaddleOCR with the D-Robotics RDK X5 dev kit.
- Best Agent System
- For the best Multi-agent Systems built on ERNIE/PaddleOCR and the CAMEL-AI framework to tackle complex real-world problems.
WHAT TO SUBMIT
The Warm-up Task offers a completion reward, while other tasks will each select 2–3 winners. In addition to the Warm-up Reward, each entrant can win only one main prize.
Warm-up Task
- The deployed web page generated from your PDF, hosted on GitHub Pages.
Model-Building Tasks
- The URL to the open-sourced fine-tuned model weights.
- The URL to the code repository containing your fine-tuning code, including training data, hyperparameters, prompt, training strategy and techniques, and a clear README file.
- A text description that explain the features and functionality of your Project.
- A demonstration video of your Project (≤5 minutes), instructional for ERNIE/PaddleOCR training and fully showcasing fine-tuned model performance.
Application-Building Tasks
- The URL to the functional demo application.
- The URL to the code repository with your full application source code / robotics code and integration, and a clear README file.
- A text description that explain the features and functionality of your Project.
- A demonstration video of your Project (≤5 minutes), showing the application scenario, key features, technical approach, and introducing the author/team.
Prizes
Warm-up Reward
Certificate and Baidu AI Studio 10 Million Tokens
Best ERNIE Fine-Tune using Unsloth
Best ERNIE Fine-Tune using LlamaFactory
Best PaddleOCR-VL Fine-Tune
Best ERNIE Multimodal Application | Powered by Baidu AI Studio
Best ERNIE Multimodal Application | Sponsored by Novita - 1st Place
$1,000 Novita Voucher and Ambassador title
Best Agent System | Sponsored by CAMEL-AI - 1st Place
1-year Eigent Pro Plan (ARV $999.90)
Best ERNIE Multimodal Application | Sponsored by Novita - 2nd Place
$700 Novita Voucher and Ambassador title
Best Agent System | Sponsored by CAMEL-AI - 2nd Place
6-month Eigent Pro Plan (ARV $499.9)
Best ERNIE Multimodal Application | Sponsored by Novita - 3rd Place
$300 Novita Voucher and Ambassador title
Best in Edge AI & Robotics | Sponsored by D-Robotics - 2nd Place
1 D-Robotics Dev Board worth about $1,000
Devpost Achievements
Submitting to this hackathon could earn you:
Judges
Jun Zhang
Baidu
Ting Liu
Baidu
Daniel Han
Unsloth
Zhangchi Feng
LLaMA-Factory
Wendong Fan
CAMEL-AI / Eigent
Viktor Hu
Novita
Lisa Li
D-Robotics
Leon Liuzx
Baidu
Yiran Ren
Baidu
Ziyu Zhang
Baidu
Youzhi Yang
Baidu
Judging Criteria
-
Application of the model
Does the project apply the models in an effective way, fitting with the categories? Can other models do the same thing, or does the project showcase the strengths of the models uniquely? -
Potential Impact
How big of an impact could the project have for its intended audience? How big of an impact could it have beyond the target community and the rest of the world? -
Creativity of the Idea
How creative and unique is the project? Does the concept exist already? If so, how much does the project improve on it? -
Documentation Quality
How well is the model/application documented for the submission? -
Demo Video Quality
Does the video clearly demonstrate the functionality and core features of the model/application, showing how it works and what problems it solves?
Questions? Email the hackathon manager
Tell your friends
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
