• NVIDIA: "Prompt Learning and Parameter Efficient Fine Tuning" with Adam Gryzwaczewski and Miguel Martínez

This talk introduces key techniques for prompt learning and parameter-efficient fine-tuning of large language models.

Despite the success of zero-shot learning capable models like ChatGPT the above approaches are orthogonal and important when fine-tuning for domain adaptation or model alignment.

This talk gives an introduction to the theory of the above approaches and follows it up with a number of hands-on examples of both prompt learning and adapter tuning using NVIDIA NeMo Framework. The second part of the talk provides an overview of how to implement efficient inference on these models, which is crucial for serving large architectures at scale.

The talk will conclude with an overview of resources that we think will be beneficial for furthering your understanding and expertise in this topic.

• Cohere: "Retrieval Augmented Generation and Its Intersection with Prompting" with David Stewart

This talk will explore different prompting techniques for accessing and leveraging your external data within an LLM.

While these models are very capable out of the box, users often have external data that they would like the model to have access to. By utilizing search and prompting, models can access this external knowledge at inference time.

This talk will walk through these retrieval and prompting techniques while also providing concrete examples.