phi3-prompt-engineer-gguf

Phi-3 Prompt Engineer (GGUF)

This repository contains GGUF quantized versions of the Phi-3 Prompt Engineer model, fine-tuned to excel at refining user requests into detailed, structured prompts for LLMs.

Model Description

• Base Model: microsoft/Phi-3-mini-4k-instruct
• Fine-tuning: LoRA adapters trained on a custom dataset of prompt refinement examples.
• Purpose: To act as a specialized "Prompt Engineer" agent, converting vague user ideas into high-quality, actionable system instructions and prompts.

Available Quantizations

The following GGUF files are available for download. Q4_K_M is recommended for most users as it balances performance and quality perfectly.

Usage

Ollama

You can use these files directly with Ollama.

1. Create a Modelfile:

   FROM ./phi3-prompt-engineer-q4_k_m.gguf
   SYSTEM "You are an Expert Prompt Engineer. Your goal is to refine the following user request into a clear, structured, and detailed prompt."
   

2. Create and run the model:

   ollama create phi3-pe -f Modelfile
   ollama run phi3-pe
   

llama.cpp

./main -m phi3-prompt-engineer-q4_k_m.gguf -p "You are an Expert Prompt Engineer. Refine this: make a snake game in python"

Training Data

The model was trained on a proprietary dataset (raw_dataset.jsonl) consisting of:

• Input: Raw, often vague user requests (e.g., "make a login page").
• Output: Detailed, structured Prompt Engineering responses including personas, constraints, and specific requirements.
• Focus: Web development, Python scripting, creative writing, and system design tasks.

License

MIT