أحدث المنشورات وبراءات الاختراع حول أمثلة على تحويل فورييه السريع (FFT)

Automatic Configuration Search for Parameter-Efficient Fine-Tuning

Published on 2024-05-25 by Han Zhou, Xingchen Wan, Ivan Vuli?, Anna Korhonen @MIT

Abstract: Large pretrained language models are widely used in downstream NLP tasks via task-specific fine-tuning, but such procedures can be costly. Recently, Parameter-Efficient Fine-Tuning (PEFT) methods have achieved strong task performance while updating much fewer parameters than full model fine-tuning (FFT). However, it is non-trivial to make informed design choices on the PEFT configurations, such as their architecture, the number of tunable parameters, and even the layers in which the PEFT modules[...]

Our summary: Efficient fine-tuning method achieved with fewer parameters. AutoPEFT uses Bayesian optimization to discover optimal configurations. Outperforms existing methods on GLUE and SuperGLUE tasks. Pareto-optimal set balances performance and cost.

automatic configuration search, parameter-efficient fine-tuning, neural architecture search, pretrained language models, task-specific fine-tuning

Publication