The open-source framework for preventing catastrophic forgetting during LLM fine-tuning with LoRA and QLoRA.
Founded by Siddharth Lakshmi Narayanan, Atharv Ranjan, and Shreyas Gandhe
PyRecall is an open-source machine learning framework focused on continual learning and reliable LLM fine-tuning. It helps developers and researchers adapt large language models to new tasks while preserving previously learned capabilities.
Unlike traditional fine-tuning approaches that can overwrite existing knowledge, PyRecall provides tools to monitor, evaluate, and reduce catastrophic forgetting during model adaptation.
Our mission is to make continual learning practical, accessible, and reproducible through high-quality open-source software.
We believe AI development should be:
- Open
- Research-driven
- Reproducible
- Accessible
- Community-focused
PyRecall is our flagship Python library for continual learning and modern fine-tuning workflows.
Current capabilities include:
- LoRA and QLoRA fine-tuning
- Built-in evaluation metrics
- Checkpoint management
- Command-line interface
- Modular architecture for research and experimentation
- Comprehensive documentation
As the organization grows, additional libraries and tools will be developed under the PyRecall ecosystem.
PyRecall is community-driven, and contributions of all sizes are welcome.
You can contribute by:
- Improving existing features
- Reporting bugs
- Proposing new ideas
- Enhancing documentation
- Writing tests
- Reviewing pull requests
If you're looking to get started, browse issues labeled good first issue.
Our goal is to build an open community around continual learning and AI engineering.
Whether you're a student, researcher, or software engineer, we welcome your contributions and feedback.
We envision an ecosystem where fine-tuning AI models is reliable, efficient, and accessible to everyone.
By building open-source tools and encouraging collaboration, we aim to accelerate research and make continual learning a standard part of modern machine learning workflows.
Each repository within the PyRecall organization includes its own license and contribution guidelines.
Please refer to the documentation of individual repositories for project-specific information.
Keep your models balanced.
