Fine Tuning and Evaluating LLMs for Agentic Use Cases

This session is focused on fine-tuning large language model (LLM) agents, acquiring crucial insights and techniques for enhancing the performance and specificity of local LLM agents in application automation. We'll explore a variety of topics, enabled by Weights & Biases, including:

• Fine-tuning techniques: Learn about LORA (low-rank adaptation) and its role in refining LLM behavior for specific applications for both open source (Llama2) and closed source models (GPT 3.5-turbo)

• Metrics and logging: Understand the importance of tracking the right metrics and maintaining detailed logs as it relates to Language Models;

• Model Checkpointing and Comparison: Implement systematic model checkpointing for tracking progress and comparing iterations, facilitating optimal version selection through detailed performance analysis.

• Practical evaluation: Engage in hands-on evaluations to assess the improvements in your fine-tuned models, both quantitatively and qualitatively