Fine-tuned microsoft/Phi-4-mini-instruct on GitHub data

Fine-tuned microsoft/Phi-4-mini-instruct on dotnet/runtime GitHub data.

Project Structure

• agent/: Contains the core implementation of the GitHub agent.
• github-crawler/: Scripts and utilities for data collection and preprocessing, including extracting issues and PRs from GitHub.
• model/: Code for fine-tuning and deploying the language model, including training scripts and model configurations.
• scripts/: Deployment and automation.

Goals

• Fine-tune a language model to understand and propose solutions for .NET runtime issues.
• Deploy the model as a GitHub agent on Azure.
• Implement a feedback loop for continuous improvement.

Quick start

1. Collect raw data from Github repository
2. Generate dataset
3. Finetune model

Demo

• Dataset: https://huggingface.co/datasets/kotlarmilos/dotnet-runtime
• Model: https://huggingface.co/kotlarmilos/dotnet-runtime
• Space: https://huggingface.co/spaces/kotlarmilos/dotnet-runtime