Self-Instruct: Aligning Language Models with Self-Generated Instructions
arXiv:2212.10560
Abstract
Large "instruction-tuned" language models (i.e., finetuned to respond to instructions) have demonstrated a remarkable ability to generalize zero-shot to new tasks. Nevertheless, they depend heavily on human-written instruction data that is often limited in quantity, diversity, and creativity, therefore hindering the generality of the tuned model. We introduce Self-Instruct, a framework for improving the instruction-following capabilities of pretrained language models by bootstrapping off their own generations. Our pipeline generates instructions, input, and output samples from a language model, then filters invalid or similar ones before using them to finetune the original model. Applying our method to the vanilla GPT3, we demonstrate a 33% absolute improvement over the original model on Super-NaturalInstructions, on par with the performance of InstructGPT-001, which was trained with private user data and human annotations. For further evaluation, we curate a set of expert-written instructions for novel tasks, and show through human evaluation that tuning GPT3 with Self-Instruct outperforms using existing public instruction datasets by a large margin, leaving only a 5% absolute gap behind InstructGPT-001. Self-Instruct provides an almost annotation-free method for aligning pre-trained language models with instructions, and we release our large synthetic dataset to facilitate future studies on instruction tuning. Our code and data are available at https://github.com/yizhongw/self-instruct.
ACL 2023 camera ready, 23 pages, 9 figures, 11 tables
Cited by in corpus (8)
- Vision-Language Models for Medical Report Generation and Visual Question Answering: A Review
- Opening up ChatGPT: Tracking openness, transparency, and accountability in instruction-tuned text generators
- HPC-GPT: Integrating Large Language Model for High-Performance Computing
- A Survey on Symbolic Knowledge Distillation of Large Language Models
- Automatic Personalized Impression Generation for PET Reports Using Large Language Models
- MMInstruct: A High-Quality Multi-Modal Instruction Tuning Dataset with Extensive Diversity
- Evaluating Language Model Agency through Negotiations
- Implicit Humanization in Everyday LLM Moral Judgments