CTRL: A Conditional Transformer Language Model for Controllable Generation

Abstract

Large-scale language models show promising text generation capabilities, but users cannot easily control particular aspects of the generated text. We release CTRL, a 1.63 billion-parameter conditional transformer language model, trained to condition on control codes that govern style, content, and task-specific behavior. Control codes were derived from structure that naturally co-occurs with raw text, preserving the advantages of unsupervised learning while providing more explicit control over text generation. These codes also allow CTRL to predict which parts of the training data are most likely given a sequence. This provides a potential method for analyzing large amounts of data via model-based source attribution. We have released multiple full-sized, pretrained versions of CTRL at https://github.com/salesforce/ctrl.

References in corpus (37)

• Adam: A Method for Stochastic Optimization
• Distributed Representations of Words and Phrases and their Compositionality
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• Natural Language Processing (almost) from Scratch
• Teaching Machines to Read and Comprehend
• Cross-lingual Language Model Pretraining
• The Curious Case of Neural Text Degeneration
• SQuAD: 100,000+ Questions for Machine Comprehension of Text
• Layer Normalization
• Generating Long Sequences with Sparse Transformers
• Semi-supervised Sequence Learning
• Regularizing and Optimizing LSTM Language Models
• SearchQA: A New Q&A Dataset Augmented with Context from a Search Engine
• Abstractive Text Summarization Using Sequence-to-Sequence RNNs and Beyond
• The Natural Language Decathlon: Multitask Learning as Question Answering
• A Simple Method for Commonsense Reasoning
• One Model To Learn Them All
• Neural Text Generation with Unlikelihood Training
• Tying Word Vectors and Word Classifiers: A Loss Framework for Language Modeling
• Fast Decoding in Sequence Models using Discrete Latent Variables
• Adafactor: Adaptive Learning Rates with Sublinear Memory Cost
• Hierarchical Neural Story Generation
• Unsupervised Question Answering by Cloze Translation
• HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering
• Unsupervised Neural Machine Translation
• Defending Against Neural Fake News
• Stochastic Gradient Methods with Layer-wise Adaptive Moments for Training of Deep Networks
• NewsQA: A Machine Comprehension Dataset
• Large Memory Layers with Product Keys
• A Joint Many-Task Model: Growing a Neural Network for Multiple NLP Tasks
• Unifying Question Answering, Text Classification, and Regression via Span Extraction
• Memory-Efficient Adaptive Optimization
• Pretrained AI Models: Performativity, Mobility, and Change
• Domain Control for Neural Machine Translation
• ELI5: Long Form Question Answering
• Answers Unite! Unsupervised Metrics for Reinforced Summarization Models
• SumQE: a BERT-based Summary Quality Estimation Model

Cited by in corpus (183)

• Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
• On the Opportunities and Risks of Foundation Models
• Pre-trained Models for Natural Language Processing: A Survey
• Evaluating Large Language Models Trained on Code
• Multilingual Denoising Pre-training for Neural Machine Translation
• Decision Transformer: Reinforcement Learning via Sequence Modeling
• Prefix-Tuning: Optimizing Continuous Prompts for Generation
• Towards a Human-like Open-Domain Chatbot
• TweepFake: about Detecting Deepfake Tweets
• A Simple Language Model for Task-Oriented Dialogue
• Plug and Play Language Models: A Simple Approach to Controlled Text Generation
• Data Augmentation using Pre-trained Transformer Models
• Large Language Models for Code: Security Hardening and Adversarial Testing
• SOLOIST: Building Task Bots at Scale with Transfer Learning and Machine Teaching
• Recipes for Safety in Open-domain Chatbots
• It's Morphin' Time! Combating Linguistic Discrimination with Inflectional Perturbations
• Survey of different Large Language Model Architectures: Trends, Benchmarks, and Challenges
• Few-Shot Text Generation with Pattern-Exploiting Training
• Abstractive Text Summarization: State of the Art, Challenges, and Improvements
• Few-Shot Named Entity Recognition: A Comprehensive Study
• Pop Music Transformer: Beat-based Modeling and Generation of Expressive Pop Piano Compositions
• Hash Layers For Large Sparse Models
• exBERT: A Visual Analysis Tool to Explore Learned Representations in Transformers Models
• Transformer-based Map Matching Model with Limited Ground-Truth Data using Transfer-Learning Approach
• Explaining Question Answering Models through Text Generation
• PlotMachines: Outline-Conditioned Generation with Dynamic Plot State Tracking
• Transformer-based Conditional Variational Autoencoder for Controllable Story Generation
• Wordcraft: a Human-AI Collaborative Editor for Story Writing
• ProGen: Language Modeling for Protein Generation
• Controlling Style in Generated Dialogue
• A Distributional Approach to Controlled Text Generation
• Few-shot Natural Language Generation for Task-Oriented Dialog
• Clinical Insights: A Comprehensive Review of Language Models in Medicine
• Artificial Text Detection via Examining the Topology of Attention Maps
• GenNI: Human-AI Collaboration for Data-Backed Text Generation
• Efficient Retrieval Augmented Generation from Unstructured Knowledge for Task-Oriented Dialog
• Controllable Semantic Parsing via Retrieval Augmentation
• Synthetic Data Augmentation for Zero-Shot Cross-Lingual Question Answering
• On Feature Normalization and Data Augmentation
• scb-mt-en-th-2020: A Large English-Thai Parallel Corpus
• Applying Transformer-based Text Summarization for Keyphrase Generation
• Controllable Data Generation by Deep Learning: A Review
• Data Movement Is All You Need: A Case Study on Optimizing Transformers
• Dungeons and Dragons as a Dialog Challenge for Artificial Intelligence
• Safety Analysis in the Era of Large Language Models: A Case Study of STPA using ChatGPT
• Neural Language Generation: Formulation, Methods, and Evaluation
• Adversarial Robustness of Neural-Statistical Features in Detection of Generative Transformers
• Say As You Wish: Fine-grained Control of Image Caption Generation with Abstract Scene Graphs
• Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models
• Compound Word Transformer: Learning to Compose Full-Song Music over Dynamic Directed Hypergraphs
• Residual Energy-Based Models for Text
• Back to the Future: Unsupervised Backprop-based Decoding for Counterfactual and Abductive Commonsense Reasoning
• C5T5: Controllable Generation of Organic Molecules with Transformers
• Anticipating Safety Issues in E2E Conversational AI: Framework and Tooling
• CoCon: A Self-Supervised Approach for Controlled Text Generation
• AraGPT2: Pre-Trained Transformer for Arabic Language Generation
• Finding Experts in Transformer Models
• PLATO: Pre-trained Dialogue Generation Model with Discrete Latent Variable
• SceneFormer: Indoor Scene Generation with Transformers
• Better Distractions: Transformer-based Distractor Generation and Multiple Choice Question Filtering
• DSGPT: Domain-Specific Generative Pre-Training of Transformers for Text Generation in E-commerce Title and Review Summarization
• Style Example-Guided Text Generation using Generative Adversarial Transformers
• Transferring Inductive Biases through Knowledge Distillation
• Mirostat: A Neural Text Decoding Algorithm that Directly Controls Perplexity
• Findings of the The RuATD Shared Task 2022 on Artificial Text Detection in Russian
• Ratatouille: A tool for Novel Recipe Generation
• MEGATRON-CNTRL: Controllable Story Generation with External Knowledge Using Large-Scale Language Models
• Deep Extrapolation for Attribute-Enhanced Generation
• Optimus: Organizing Sentences via Pre-trained Modeling of a Latent Space
• DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts
• Detoxifying Language Models Risks Marginalizing Minority Voices
• Zero-Shot Fine-Grained Style Transfer: Leveraging Distributed Continuous Style Representations to Transfer To Unseen Styles
• Towards Controllable Biases in Language Generation
• TLDR: Token Loss Dynamic Reweighting for Reducing Repetitive Utterance Generation
• WangchanBERTa: Pretraining transformer-based Thai Language Models
• Unsupervised Contextual Paraphrase Generation using Lexical Control and Reinforcement Learning
• Unsupervised Editing for Counterfactual Stories
• POINTER: Constrained Progressive Text Generation via Insertion-based Generative Pre-training
• PCAE: A Framework of Plug-in Conditional Auto-Encoder for Controllable Text Generation
• Conditioned Natural Language Generation using only Unconditioned Language Model: An Exploration
• ETC-NLG: End-to-end Topic-Conditioned Natural Language Generation
• CBAG: Conditional Biomedical Abstract Generation
• Improved Natural Language Generation via Loss Truncation
• The Offense-Defense Balance of Scientific Knowledge: Does Publishing AI Research Reduce Misuse?
• Keyphrase Generation with Cross-Document Attention
• RADDLE: An Evaluation Benchmark and Analysis Platform for Robust Task-oriented Dialog Systems
• Multimodal Conditionality for Natural Language Generation
• You are right. I am ALARMED -- But by Climate Change Counter Movement
• Human Language Modeling
• The Rumour Mill: Making the Spread of Misinformation Explicit and Tangible
• PopMAG: Pop Music Accompaniment Generation
• DAGA: Data Augmentation with a Generation Approach for Low-resource Tagging Tasks
• Controllable Abstractive Dialogue Summarization with Sketch Supervision
• SideControl: Controlled Open-domain Dialogue Generation via Additive Side Networks
• Decentralized Attribution of Generative Models
• Template Guided Text Generation for Task-Oriented Dialogue
• Conditional Text Generation for Harmonious Human-Machine Interaction
• What's New? Summarizing Contributions in Scientific Literature
• RSTGen: Imbuing Fine-Grained Interpretable Control into Long-FormText Generators
• Few-shot Sequence Learning with Transformers
• Composed Variational Natural Language Generation for Few-shot Intents
• Which *BERT? A Survey Organizing Contextualized Encoders
• Outline to Story: Fine-grained Controllable Story Generation from Cascaded Events
• KGPT: Knowledge-Grounded Pre-Training for Data-to-Text Generation
• Language Model Augmented Relevance Score
• GeDi: Generative Discriminator Guided Sequence Generation
• Conditional Augmentation for Aspect Term Extraction via Masked Sequence-to-Sequence Generation
• PatentTransformer-2: Controlling Patent Text Generation by Structural Metadata
• Neural Rule-Execution Tracking Machine For Transformer-Based Text Generation
• Reward-Augmented Decoding: Efficient Controlled Text Generation With a Unidirectional Reward Model
• What Changes Can Large-scale Language Models Bring? Intensive Study on HyperCLOVA: Billions-scale Korean Generative Pretrained Transformers
• Nano: Nested Human-in-the-Loop Reward Learning for Few-shot Language Model Control
• Pre-train and Plug-in: Flexible Conditional Text Generation with Variational Auto-Encoders
• Conditional Neural Generation using Sub-Aspect Functions for Extractive News Summarization
• Reverse Engineering Configurations of Neural Text Generation Models
• Of Human Criteria and Automatic Metrics: A Benchmark of the Evaluation of Story Generation
• Staircase Attention for Recurrent Processing of Sequences
• DocNLI: A Large-scale Dataset for Document-level Natural Language Inference
• Collocation2Text: Controllable Text Generation from Guide Phrases in Russian
• Directed Beam Search: Plug-and-Play Lexically Constrained Language Generation
• Generating Synthetic Text Data to Evaluate Causal Inference Methods
• COBE: Contextualized Object Embeddings from Narrated Instructional Video
• Changing the Mind of Transformers for Topically-Controllable Language Generation
• Argumentative Text Generation in Economic Domain
• Controlled Text Generation as Continuous Optimization with Multiple Constraints
• Language Model Evaluation in Open-ended Text Generation
• Investigation of Sentiment Controllable Chatbot
• PAIR: Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation
• Towards Generating Citation Sentences for Multiple References with Intent Control
• Which Discriminator for Cooperative Text Generation?
• Melody Structure Transfer Network: Generating Music with Separable Self-Attention
• Improving Diversity of Neural Text Generation via Inverse Probability Weighting
• You Can Do Better! If You Elaborate the Reason When Making Prediction
• Adapting a Language Model for Controlled Affective Text Generation
• Emergent Properties of Finetuned Language Representation Models
• Generative Models for Security: Attacks, Defenses, and Opportunities
• TextSETTR: Few-Shot Text Style Extraction and Tunable Targeted Restyling
• Self-conditioning pre-trained language models
• Graphmax for Text Generation
• Goal-driven text descriptions for images
• Math Word Problem Generation with Mathematical Consistency and Problem Context Constraints
• Towards Universality in Multilingual Text Rewriting
• Increasing Faithfulness in Knowledge-Grounded Dialogue with Controllable Features
• Controllable Dialogue Generation with Disentangled Multi-grained Style Specification and Attribute Consistency Reward
• A New Approach to Overgenerating and Scoring Abstractive Summaries
• Global-aware Beam Search for Neural Abstractive Summarization
• Zero-Shot Controlled Generation with Encoder-Decoder Transformers
• Plan ahead: Self-Supervised Text Planning for Paragraph Completion Task
• IGA : An Intent-Guided Authoring Assistant
• Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes
• DYPLOC: Dynamic Planning of Content Using Mixed Language Models for Text Generation
• Linguacodus: A Synergistic Framework for Transformative Code Generation in Machine Learning Pipelines
• AI-generated text boundary detection with RoFT
• CFL: Causally Fair Language Models Through Token-level Attribute Controlled Generation
• Multi-Pair Text Style Transfer on Unbalanced Data
• Generating abstractive summaries of Lithuanian news articles using a transformer model
• Domain Conditional Predictors for Domain Adaptation
• CO-STAR: Conceptualisation of Stereotypes for Analysis and Reasoning
• Realistic simulation of users for IT systems in cyber ranges
• MEDCOD: A Medically-Accurate, Emotive, Diverse, and Controllable Dialog System
• Bag-of-Vectors Autoencoders for Unsupervised Conditional Text Generation
• Logic-Consistency Text Generation from Semantic Parses
• Classification as Decoder: Trading Flexibility for Control in Medical Dialogue
• Towards Explainable Fact Checking
• SHARE: a System for Hierarchical Assistive Recipe Editing
• Neural Path Hunter: Reducing Hallucination in Dialogue Systems via Path Grounding
• InFillmore: Frame-Guided Language Generation with Bidirectional Context
• Attribute Alignment: Controlling Text Generation from Pre-trained Language Models
• Automatically Exposing Problems with Neural Dialog Models
• Efficient Attribute Injection for Pretrained Language Models
• Style Control for Schema-Guided Natural Language Generation
• Beyond [CLS] through Ranking by Generation
• Metadata Shaping: Natural Language Annotations for the Tail
• Char2Subword: Extending the Subword Embedding Space Using Robust Character Compositionality
• Facilitating the Communication of Politeness through Fine-Grained Paraphrasing
• WakaVT: A Sequential Variational Transformer for Waka Generation
• "TL;DR:" Out-of-Context Adversarial Text Summarization and Hashtag Recommendation
• Estimating Subjective Crowd-Evaluations as an Additional Objective to Improve Natural Language Generation
• Semantic Frame Forecast
• Exposure Bias versus Self-Recovery: Are Distortions Really Incremental for Autoregressive Text Generation?
• LogicalFactChecker: Leveraging Logical Operations for Fact Checking with Graph Module Network
• Examining the rhetorical capacities of neural language models
• Fingerprinting Fine-tuned Language Models in the Wild