Master LLMs with our FREE course in collaboration with Activeloop & Intel Disruptor Initiative. Join now!

Publication

Parameter Efficient Fine Tuning (PEFT), Inference, and Evaluation of LLM model using LoRA
Latest   Machine Learning

Parameter Efficient Fine Tuning (PEFT), Inference, and Evaluation of LLM model using LoRA

Last Updated on May 9, 2024 by Editorial Team

Author(s): Steve George

Originally published on Towards AI.

LLM has been able to solve a lot of real world problems which includes sentimental analysis, classification, summarization, chatbot and many more. However these models needs to be retrained based on the use-case to improve the model accuracy.

Below are the problems of fine-tuning a modela. Compute: To train a large LLM model, a huge compute is required, as the number of weights are significantly on the higher side.b. Fine-tuned model size: The fine-tuned model size gets larger and larger as we keep training.c. Cost: The above factors, in turn, will increase the cost of managing the model.

Note: Catastrophic Forgetting is a problem faced by many models. Suppose a model is trained on task A and is currently being training on task B, there is a possibility where the model forgets how to perform on task A. This is because the model weights gets updated while training on task B, resulting in poor performance. Fine-tuning of the model may cause Catastrophic Forgetting.

PEFT fine-tunes a small number of model parameters while freezing most parameters of the LLM model. By doing so, we can have the following;a. Reduce cost and storageb. Improve development speedc. Reduce the probability of Catastrophic Forgetting

PEFT Training

In PEFT, lets… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI

Feedback ↓