The goal of this work is to align an open-source large-language model to the intent of the user. Throughout the work we assume access to a larger teacher model... Our goal is to produce a student model πθ and our approach follows similar stages as InstructGPT.
Distilled Supervised Fine-Tuning (dSFT) involves training a raw LLM to respond to user prompts through supervised fine-tuning with a dataset of high-quality instructions and responses or by having the model generate instructions and responses to train directly. Self-instruct protocol is followed for dSFT.
#large-language-model #supervised-fine-tuning #distilled-supervised-fine-tuning #user-intent-alignment
Collection
[
|
...
]