- 🎓 I’m currently a PhD student in CS at EPFL Data Science Lab
- 🌱 My research interest is reliable and controllable LLM generation via constrained decoding
- 🤗 I contribute to the open-source community in my spare time.
- 🔬 I’m always looking for motivated master's students interested in my research
- 📫 How to reach me: if you want to work with me on a semester project or master thesis, have a look at openings and feel free to drop an email
- zip2zip: Inference-time adaptive vocabularies for language models via token compression, enabling faster inference by dynamically adjusting tokenization with hypertokens.
- zip2zip-compression: Core LZW-based compression and hypertoken encoding utilities powering the zip2zip framework.
- Transformers-CFG: A library to extend Huggingface Transformers with constrained decoding methods for LLM
- Awesome-LLM-Constrained-Decoding: A curated list of papers and resources on constrained decoding methods for LLM
- JsonSchemaBench: The first systematic investigation into the performance of constrained decoding frameworks for LLM, alo available on EleutherAI/lm-evaluation-harness
- JsonSchemaBench-Leaderboard: Tracking the evolution of LLM capacity on structured generation.