Louis Martin

Titre

"Finetuning LLMs, From Llama to Mistral"

Résumé

"The presentation by Louis Martin from Mistral AI focuses on the fine-tuning of large language models, specifically detailing the processes and advancements from Llama to Mistral. It covers essential aspects such as data collection, alignment pipelines, and the use of techniques like rejection sampling and Direct Preference Optimization (DPO) to enhance model performance and alignment with human preferences."

Bio

"Louis Martin, a Research Scientist at Mistral AI, leads the post-training efforts for large language models (LLMs). He has significantly contributed to the development and fine-tuning of models like Llama 2 and Llama 3, focusing on techniques such as supervised fine-tuning and direct preference optimization to align these models with human preferences. His work also explores the impact of training data on language modeling, particularly for French language models."