Large language model enhanced with prompt-based vanilla distillation for sentence embeddings

In this dissertation, the prompt-based method PromptEOL is used to train the opt- 2.7b model with the Parameter-Efficient Fine-Tuning method to reduce the number of training parameters and GPU memory usage. Then the opt-2.7b-lora model is used as the teacher model to train the student model under...

Full description

Saved in:
Bibliographic Details
Main Author: Wang, Minghao
Other Authors: Lihui Chen
Format: Thesis-Master by Coursework
Language:English
Published: Nanyang Technological University 2024
Subjects:
Online Access:https://hdl.handle.net/10356/173839
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:In this dissertation, the prompt-based method PromptEOL is used to train the opt- 2.7b model with the Parameter-Efficient Fine-Tuning method to reduce the number of training parameters and GPU memory usage. Then the opt-2.7b-lora model is used as the teacher model to train the student model under the distillation framework of DistillCSE with the vanilla distillation. The core method of evaluation we use centers on Semantic Textual Similarity detection.