Scaling Instruction-Finetuned Language Models

Abstract

Finetuning language models on a collection of datasets phrased as instructions has been shown to improve model performance and generalization to unseen tasks. In this paper we explore instruction finetuning with a particular focus on (1) scaling the number of tasks, (2) scaling the model size, and (3) finetuning on chain-of-thought data. We find that instruction finetuning with the above aspects dramatically improves performance on a variety of model classes (PaLM, T5, U-PaLM), prompting setups (zero-shot, few-shot, CoT), and evaluation benchmarks (MMLU, BBH, TyDiQA, MGSM, open-ended generation). For instance, Flan-PaLM 540B instruction-finetuned on 1.8K tasks outperforms PALM 540B by a large margin (+9.4% on average). Flan-PaLM 540B achieves state-of-the-art performance on several benchmarks, such as 75.2% on five-shot MMLU. We also publicly release Flan-T5 checkpoints, which achieve strong few-shot performance even compared to much larger models, such as PaLM 62B. Overall, instruction finetuning is a general method for improving the performance and usability of pretrained language models.

arXiv commentPublic checkpoints: https://huggingface.co/docs/transformers/model_doc/flan-t5

Cited by in corpus (88)

• Adapted Large Language Models Can Outperform Medical Experts in Clinical Text Summarization
• A Survey on Multimodal Large Language Models
• Opportunities and Challenges for ChatGPT and Large Language Models in Biomedicine and Health
• Large language models in medicine: the potentials and pitfalls
• Power Hungry Processing: Watts Driving the Cost of AI Deployment?
• Large AI Models in Health Informatics: Applications, Challenges, and the Future
• Uncovering ChatGPT's Capabilities in Recommender Systems
• Generative AI for Synthetic Data Across Multiple Medical Modalities: A Systematic Review of Recent Developments and Challenges
• ChartGPT: Leveraging LLMs to Generate Charts from Abstract Natural Language
• LLMParser: An Exploratory Study on Using Large Language Models for Log Parsing
• Deception Abilities Emerged in Large Language Models
• Construction of Knowledge Graphs: State and Challenges
• Large Language Models as Zero-Shot Human Models for Human-Robot Interaction
• Fine Tuning vs. Retrieval Augmented Generation for Less Popular Knowledge
• Socially Aware Synthetic Data Generation for Suicidal Ideation Detection Using Large Language Models
• Language Model Crossover: Variation through Few-Shot Prompting
• Neural Natural Language Processing for Long Texts: A Survey on Classification and Summarization
• One-Shot Labeling for Automatic Relevance Estimation
• The Information Retrieval Experiment Platform
• ThoughtSource: A central hub for large language model reasoning data
• A Survey on Symbolic Knowledge Distillation of Large Language Models
• A Survey on Occupancy Perception for Autonomous Driving: The Information Fusion Perspective
• Exploring the Roles of Large Language Models in Reshaping Transportation Systems: A Survey, Framework, and Roadmap
• Examining Inter-Consistency of Large Language Models Collaboration: An In-depth Analysis via Debate
• Do LLMs Understand Social Knowledge? Evaluating the Sociability of Large Language Models with SocKET Benchmark
• Towards Explainable Conversational Recommender Systems
• Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties
• Large-Scale Text Analysis Using Generative Language Models: A Case Study in Discovering Public Value Expressions in AI Patents
• Prompted LLMs as Chatbot Modules for Long Open-domain Conversation
• Clinical Insights: A Comprehensive Review of Language Models in Medicine
• Applications of Generative AI in Healthcare: algorithmic, ethical, legal and societal considerations
• Inseq: An Interpretability Toolkit for Sequence Generation Models
• Large language models for automated scholarly paper review: A survey
• Opportunities for Large Language Models and Discourse in Engineering Design
• Reasoning Beyond Limits: Advances and Open Problems for LLMs
• DeTiME: Diffusion-Enhanced Topic Modeling using Encoder-decoder based LLM
• BLP-2023 Task 2: Sentiment Analysis
• Retrieving Evidence from EHRs with LLMs: Possibilities and Challenges
• Utilizing Cognitive Signals Generated during Human Reading to Enhance Keyphrase Extraction from Microblogs
• Automatically Generating UI Code from Screenshot: A Divide-and-Conquer-Based Approach
• Federated Fine-Tuning of LLMs on the Very Edge: The Good, the Bad, the Ugly
• A survey of textual cyber abuse detection using cutting-edge language models and large language models
• Health Text Simplification: An Annotated Corpus for Digestive Cancer Education and Novel Strategies for Reinforcement Learning
• Discrete Prompt Compression with Reinforcement Learning
• XTREME-UP: A User-Centric Scarce-Data Benchmark for Under-Represented Languages
• Comparison of Large Language Models for Generating Contextually Relevant Questions
• Unveiling the Vulnerability of Private Fine-Tuning in Split-Based Frameworks for Large Language Models: A Bidirectionally Enhanced Attack
• Jointly Extracting Interventions, Outcomes, and Findings from RCT Reports with LLMs
• Unified Modeling Language Code Generation from Diagram Images Using Multimodal Large Language Models
• Prompting Video-Language Foundation Models with Domain-specific Fine-grained Heuristics for Video Question Answering
• A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis
• Democratizing LLMs: An Exploration of Cost-Performance Trade-offs in Self-Refined Open-Source Models
• Learning to Paraphrase Sentences to Different Complexity Levels
• On the Potential and Limitations of Few-Shot In-Context Learning to Generate Metamorphic Specifications for Tax Preparation Software
• Mirror: A Natural Language Interface for Data Querying, Summarization, and Visualization
• One Size Does Not Fit All: Investigating Efficacy of Perplexity in Detecting LLM-Generated Code
• Fast and Accurate Factual Inconsistency Detection Over Long Documents
• Writing Style Matters: An Examination of Bias and Fairness in Information Retrieval Systems
• Emotion-aware Dual Cross-Attentive Neural Network with Label Fusion for Stance Detection in Misinformative Social Media Content
• Knowledge Graph Based Repository-Level Code Generation
• OPT-R: Exploring the Role of Explanations in Finetuning and Prompting for Reasoning Skills of Large Language Models
• Blacks is to Anger as Whites is to Joy? Understanding Latent Affective Bias in Large Pre-trained Neural Language Models
• Optimal strategies to perform multilingual analysis of social content for a novel dataset in the tourism domain
• Learning to Taste: A Multimodal Wine Dataset
• HELIOT: LLM-Based CDSS for Adverse Drug Reaction Management
• Deep Natural Language Feature Learning for Interpretable Prediction
• FlanEC: Exploring Flan-T5 for Post-ASR Error Correction
• When lies are mostly truthful: automated verbal deception detection for embedded lies
• Unsupervised Text Embedding Space Generation Using Generative Adversarial Networks for Text Synthesis
• A Scalable and Efficient Signal Integration System for Job Matching
• Parameter Efficient Diverse Paraphrase Generation Using Sequence-Level Knowledge Distillation
• Exploring the Potential of Machine Translation for Generating Named Entity Datasets: A Case Study between Persian and English
• Diable: Efficient Dialogue State Tracking as Operations on Tables
• Divergences between Language Models and Human Brains
• Large Language Models are biased to overestimate profoundness
• Argus: Leveraging Multiview Images for Improved 3-D Scene Understanding With Large Language Models
• Model-Generated Pretraining Signals Improves Zero-Shot Generalization of Text-to-Text Transformers
• What Comes Next? Evaluating Uncertainty in Neural Text Generators Against Human Production Variability
• InsightNet: Structured Insight Mining from Customer Feedback
• Exploring news intent and its application: A theory-driven approach
• MSG-Chart: Multimodal Scene Graph for ChartQA
• Automatic Generation of Highlights for Academic Paper Via Prompt-based Learning
• ModuLoRA: Finetuning 2-Bit LLMs on Consumer GPUs by Integrating with Modular Quantizers
• STEER: Semantic Turn Extension-Expansion Recognition for Voice Assistants
• Can Small Language Models Handle Context-Summarized Multi-Turn Customer-Service QA? A Synthetic Data-Driven Comparative Evaluation
• A Roadmap for Tamed Interactions with Large Language Models
• CoServe: Efficient Collaboration-of-Experts (CoE) Model Inference with Limited Memory
• Simple and Effective Input Reformulations for Translation