Google

Large Language Models can be adapted for specific tasks using several distinct methods. These can be broadly categorized into approaches that add new, trainable parameters and those that modify existing ones. The former includes techniques like using soft prompts as prefixes for each layer (Prefix Tuning) or as inputs to the embedding layer (Prompt Tuning). The latter involves fine-tuning select parts of the original model or adding and training separate, lightweight 'adapter' modules.

Classification of LLM Adaptation Methods

A research team is adapting a large language model for a specialized task. To minimize computational requirements and avoid altering the model's core knowledge, they decide to freeze all of the original model's parameters. They then introduce and train small, new neural network modules inserted between the model's existing layers. Based on this description, how is this adaptation method categorized?

Match each Large Language Model adaptation technique with the statement that best describes its core mechanism for adjusting the model.

Based on the company's constraints described in the case study, which of the two broad categories of adaptation methods (modifying existing parameters vs. adding new, trainable parameters) would be the more appropriate choice? Justify your reasoning by explaining how your chosen category addresses both of the company's constraints.

Learn Before

Related