Get Ideas and Inspiration Here

Fine-tuning Ideas

Explore fine-tuning in diverse domains such as low-resource languages, finance and economics, literature and creative writing, civil and architectural engineering, computer science, or mechanical manufacturing. If your interests lie in other niche or unique scenarios, we welcome those ideas too!

Application & Robotics Ideas

Explore innovative applications across diverse real-world scenarios such as lifestyle and entertainment, culture and media, e-commerce marketing, smart education, learning, and office productivity, and public welfare. Additionally, the exploration of robotics is welcomed, focusing both on the intelligent transformation of existing systems and the development of novel smart interactive devices.

There’s no single “correct” approach—just pick a direction you’re excited about and innovate from there!

Tools and Technologies

You can either use ERNIE or PaddleOCR-VL open-source models (free but requiring sufficient GPU resources on the device), or access them via APIs through Baidu AI Studio or Novita AI.

• Resource offered by Baidu AI Studio:
  • Baidu AI Studio offers 1 million free tokens upon login, and an additional 1million tokens after completing profile verification, plus 8 compute points per day for running projects on it.
  • Register here: https://aistudio.baidu.com/overview?lang=en (recommended to log in through GitHub or Google)
  • If you need additional GPU resources or API tokens, apply here: https://forms.gle/7fyZtBF9LoR9hmuXA

• Resource offered by Novita AI:
  • Novita AI offers $25 in credits to every participant, you can use sandbox, GPU and LLM to build your project. 
  • Register here: https://novita.ai/ernie

• Resource offered by D-Robotics

• • D-Robotics offers 10 free RDK X5 development kits to eligible participants or teams competing in the Edge AI & Robotics task—availability is limited and on a first-come, first-served basis.
  • Apply here: https://forms.gle/No78STpmr6pWJ4Sd8

• For tasks that require the use of sponsor tools, corresponding links and practical tutorials are also available via links below.

Tutorials

Warm-up Task

• Doc2Page via PaddleOCR: https://huggingface.co/spaces/PaddlePaddle/doc2page

Model-Building Tasks

• PaddleOCR-VL Fine-Tune using ERNIEKit:
  • Github: https://github.com/PaddlePaddle/ERNIE/blob/release/v1.4/docs/paddleocr_vl_sft.md
  • Google Colab: https://colab.research.google.com/drive/1yjbH1zbvBlyUq1wz0pohPaWKHO6szAXb?usp=sharing
• ERNIE Fine-Tune using LLaMA-Factory:
  • ERNIE-VL: https://github.com/hiyouga/LLaMA-Factory/pull/9521
  • ERNIE Text: https://github.com/hiyouga/LLaMA-Factory/pull/9165
• ERNIE Fine-Tune using Unsloth: https://github.com/unslothai/notebooks/pull/139
  • Google Colab:
    • ERNIE-VL: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/ERNIE_4_5_VL_28B_A3B_PT_Vision.ipynb
    • ERNIE Text: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/ERNIE_4_5_21B_A3B_PT-Conversational.ipynb
    • PaddleOCR: https://colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Paddle_OCR_(1B)_Vision.ipynb
  • AMD:
    
    • ERNIE-VL: https://oneclickamd.ai/github/unslothai/notebooks/blob/main/nb/ERNIE_4_5_VL_28B_A3B_PT_Vision.ipynb
    • ERNIE Text: https://oneclickamd.ai/github/unslothai/notebooks/blob/main/nb/ERNIE_4_5_21B_A3B_PT-Conversational.ipynb

Application-Building Tasks

• Baidu AI Studio API Doc:
  • Run ERNIE: https://ai.baidu.com/ai-doc/AISTUDIO/Mmhslv9lf
  • Run PaddleOCR-VL: https://ai.baidu.com/ai-doc/AISTUDIO/Dmh4onssk
• Novita AI API Doc:
  • Run ERNIE: https://blogs.novita.ai/how-to-access-ernie-4-5-effortless-ways-via-web-api-and-code/
  • Run PaddleOCR-VL: https://blogs.novita.ai/paddleocr-on-novita-ai/
• D-Robotics RDK:
  • Run PaddleOCR: https://github.com/D-Robotics/rdk_model_zoo/tree/main/demos/OCR/PaddleOCR

• CAMEL-AI
  • Run ERNIE: https://github.com/camel-ai/camel/blob/master/examples/models/ernie_model_example.py