glerzing Goto Github PK

5 years of experience developing C++ software and Python tests for safety-critical equipments, including 4 years in Thales.

Past experience with programming exercises on Euler Project (problems involving mathematics and programming, 205 problems solved) and CryptoHack (CTFs involving cryptography and programming).

Contributed to Tournesol, a website that recommends high-quality YouTube videos, and collects data on people's ethical preferences :

• Backend development
• Developed a search bar algorithm integrated with PostgreSQL. It was hard to find the right design and syntax, but definitely a valuable experience in NLP and SQL.
• Coauthor of Tournesol: Permissionless Collaborative Algorithmic Governance with Security Guarantees, a paper establishing a fair and secure voting scheme notably using linear algebra and Bayesian statistics.
• Tested multiple techniques for topic classification on videos, such as lbl2vec, Latent Dirichlet Allocation and LLM prompt tuning.

Integrated peft (parameter-efficient fine-tuning, e.g. Lora) in trlx, a Python library for fine-tuning large language models with reinforcement learning. This was a very interesting project for learning, which gave me some practical experience with LLMs, Hugging Face's transformers library, PyTorch, reinforcement learning, and distributed computing. I also helped on TransformerLens, an interpretability Python library using PyTorch that helps analyze the internals of transformer models.