MePO: Lightweight Prompt Optimization Model

MePO is a 7.6 billion parameter model developed by zixiaozhu, based on the Qwen2.5-7B-Instruct architecture. It is specifically fine-tuned for prompt optimization, aiming to improve the effectiveness of prompts, particularly in scenarios with limited LLM resources. The model's core function is to take an initial "Silver Prompt" and, optionally, a "Sliver Response" and a desired "Golden Response," then generate an optimized "Golden Prompt" that yields more accurate results while remaining faithful to the original factual information.

Key Capabilities

• Prompt Optimization: Transforms sub-optimal prompts into more effective ones for better AI responses.
• Low-Resource LLM Enhancement: Designed to boost prompt performance in environments where LLM resources are constrained.
• Locally Deployable: A lightweight model suitable for local deployment and research.
• Factual Fidelity: Ensures that optimized prompts strictly adhere to the factual information present in the original prompt.

Datasets and Resources

The model was trained using specialized datasets for prompt optimization, including:

• MePO
• MePO_BPO (Optimized prompts based on the BPO dataset)
• MePO_Alpaca (Optimized prompts based on the Alpaca dataset)

Further details on implementation and training scripts are available on the GitHub repository.

Good for

• Researchers and developers focused on prompt engineering and optimization techniques.
• Applications requiring improved prompt effectiveness for low-resource LLMs.
• Experimentation with automated prompt refinement to achieve more accurate and relevant AI outputs.