Prefix-Tuning: Optimizing Continuous Prompts for Generation

Abstract

Fine-tuning is the de facto way to leverage large pretrained language models to perform downstream tasks. However, it modifies all the language model parameters and therefore necessitates storing a full copy for each task. In this paper, we propose prefix-tuning, a lightweight alternative to fine-tuning for natural language generation tasks, which keeps language model parameters frozen, but optimizes a small continuous task-specific vector (called the prefix). Prefix-tuning draws inspiration from prompting, allowing subsequent tokens to attend to this prefix as if it were "virtual tokens". We apply prefix-tuning to GPT-2 for table-to-text generation and to BART for summarization. We find that by learning only 0.1\% of the parameters, prefix-tuning obtains comparable performance in the full data setting, outperforms fine-tuning in low-data settings, and extrapolates better to examples with topics unseen during training.

References in corpus (6)

• Decoupled Weight Decay Regularization
• Language Models are Few-Shot Learners
• CTRL: A Conditional Transformer Language Model for Controllable Generation
• Multilingual Denoising Pre-training for Neural Machine Translation
• Incorporating BERT into Neural Machine Translation
• Conditioned Natural Language Generation using only Unconditioned Language Model: An Exploration

Cited by in corpus (55)

• Learning to Prompt for Vision-Language Models
• LoRA: Low-Rank Adaptation of Large Language Models
• On the Opportunities and Risks of Foundation Models
• Pre-trained Models for Natural Language Processing: A Survey
• BARTScore: Evaluating Generated Text as Text Generation
• P-Tuning v2: Prompt Tuning Can Be Comparable to Fine-tuning Universally Across Scales and Tasks
• Vision-Language Models for Medical Report Generation and Visual Question Answering: A Review
• Unifying Vision-and-Language Tasks via Text Generation
• Deep Learning Approaches on Image Captioning: A Review
• Recent Advances in Natural Language Processing via Large Pre-Trained Language Models: A Survey
• Materials science in the era of large language models: a perspective
• CPT: Colorful Prompt Tuning for Pre-trained Vision-Language Models
• Multimodal Few-Shot Learning with Frozen Language Models
• Survey of different Large Language Model Architectures: Trends, Benchmarks, and Challenges
• Compacter: Efficient Low-Rank Hypercomplex Adapter Layers
• Differentiable Prompt Makes Pre-trained Language Models Better Few-shot Learners
• Large Language Models Can Be Strong Differentially Private Learners
• FVP: Fourier Visual Prompting for Source-Free Unsupervised Domain Adaptation of Medical Image Segmentation
• SentiPrompt: Sentiment Knowledge Enhanced Prompt-Tuning for Aspect-Based Sentiment Analysis
• Few-Shot Bot: Prompt-Based Learning for Dialogue Systems
• AMMUS : A Survey of Transformer-based Pretrained Models in Natural Language Processing
• Wordcraft: a Human-AI Collaborative Editor for Story Writing
• LFPT5: A Unified Framework for Lifelong Few-shot Language Learning Based on Prompt Tuning of T5
• Why Do Pretrained Language Models Help in Downstream Tasks? An Analysis of Head and Prompt Tuning
• HTLM: Hyper-Text Pre-Training and Prompting of Language Models
• Federated Few-Shot Learning for Mobile NLP
• Program Synthesis with Large Language Models
• PTR: Prompt Tuning with Rules for Text Classification
• Parameter-Efficient Transfer Learning with Diff Pruning
• Factual Probing Is [MASK]: Learning vs. Learning to Recall
• Efficient (Soft) Q-Learning for Text Generation with Limited Good Data
• Response Generation with Context-Aware Prompt Learning
• LMSanitator: Defending Prompt-Tuning Against Task-Agnostic Backdoors
• ClipCap: CLIP Prefix for Image Captioning
• Semi-Siamese Bi-encoder Neural Ranking Model Using Lightweight Fine-Tuning
• Knowledgeable or Educated Guess? Revisiting Language Models as Knowledge Bases
• Prompting Video-Language Foundation Models with Domain-specific Fine-grained Heuristics for Video Question Answering
• AdapterEM: Pre-trained Language Model Adaptation for Generalized Entity Matching using Adapter-tuning
• LAMPAT: Low-Rank Adaption for Multilingual Paraphrasing Using Adversarial Training
• Accurate, yet inconsistent? Consistency Analysis on Language Understanding Models
• Improving the Robustness of Transformer-based Large Language Models with Dynamic Attention
• A Knowledge-enhanced Two-stage Generative Framework for Medical Dialogue Information Extraction
• Training Neural Networks with Fixed Sparse Masks
• Constrained Language Models Yield Few-Shot Semantic Parsers
• DSEE: Dually Sparsity-embedded Efficient Tuning of Pre-trained Language Models
• Zero-Shot Dialogue State Tracking via Cross-Task Transfer
• Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes
• Parameter-Efficient Methods for Metastases Detection from Clinical Notes
• Composed Fine-Tuning: Freezing Pre-Trained Denoising Autoencoders for Improved Generalization
• Algorithm to Compilation Co-design: An Integrated View of Neural Network Sparsity
• Go Forth and Prosper: Language Modeling with Ancient Textual History
• Automatic Generation of Highlights for Academic Paper Via Prompt-based Learning
• Teaching Autoregressive Language Models Complex Tasks By Demonstration
• Do Language Models Perform Generalizable Commonsense Inference?
• Efficient Test Time Adapter Ensembling for Low-resource Language Varieties