LLM Evaluation: Auditing Fine-Tuned LLMs for Guaranteed Output Quality

Explore innovative techniques for evaluating and improving fine-tuned Large Language Models (LLMs) in this 33-minute conference talk by Mirakl data scientists Loic Pauletto and Pierre Lourdelet. Delve into the challenges of information retrieval from E-commerce product data sheets and learn how Mirakl developed a solution using fine-tuned LLMs. Discover qualitative evaluation methods, including language model quality metrics and hallucination detection. Understand how to leverage MLflow for automating LLM evaluation and monitoring. Gain insights into iterative quality improvement strategies through prompt engineering and dataset refinement. Learn how these methods enable rapid iteration on prompts and fine-tuned models to achieve production-level trustworthiness. Access additional resources such as the LLM Compact Guide and Big Book of MLOps to further expand your knowledge in this field.