Hard Prompts Made Easy: Gradient-Based Discrete Optimization for Prompt Tuning and Discovery

2023/02/07 by Yuxin Wen, Wen, Yuxin, Neel Jain +9 · 1 voice · 38 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Computational Physics and Python Applications #Gene expression and cancer classification #cs.CL #cs.LG

paper · pdf · doi:10.48550/arxiv.2302.03668

openalex publication_date 2023/02/07 · openalex created_date 2023/02/10 · openalex updated_date 2026/07/28

Abstract

The strength of modern generative models lies in their ability to be controlled through text-based prompts. Typical "hard" prompts are made from interpretable words and tokens, and must be hand-crafted by humans. There are also "soft" prompts, which consist of continuous feature vectors. These can be discovered using powerful optimization methods, but they cannot be easily interpreted, re-used across models, or plugged into a text-based interface. We describe an approach to robustly optimize hard text prompts through efficient gradient-based optimization. Our approach automatically generates hard text-based prompts for both text-to-image and text-to-text applications. In the text-to-image setting, the method creates hard prompts for diffusion models, allowing API users to easily generate, discover, and mix and match image concepts without prior knowledge on how to prompt the model. In the text-to-text setting, we show that hard prompts can be automatically discovered that are effective in tuning LMs for classification.

Cited by

• Language Models are Injective and Hence Invertible
• PRISM: Prompt Refinement via Image-grounded Self-rewarding Mechanism for Text-to-Image Generation
• GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs
• Open Ad-hoc Categorization with Contextualized Feature Learning
• Exposing Hidden Biases in Text-to-Image Models via Automated Prompt Search
• In-Context Representation Hijacking
• PROMPTMINER: Black-Box Prompt Stealing against Text-to-Image Generative Models via Reinforcement Learning and Fuzz Optimization
• CAHS-Attack: CLIP-Aware Heuristic Search Attack Method for Stable Diffusion
• TASO: Jailbreak LLMs via Alternative Template and Suffix Optimization
• Transformer Injectivity & Geometric Robustness - Analytic Margins and Bi-Lipschitz Uniformity of Sequence-Level Hidden States
• Prompt-R1: Collaborative Automatic Prompting Framework via End-to-end Reinforcement Learning
• EraseFlow: Learning Concept Erasure Policies via GFlowNet-Driven Alignment
• Reimagining Safety Alignment with An Image
• From the Rock Floor to the Cloud: A Systematic Survey of State-of-the-Art NLP in Battery Life Cycle
• Rethinking Robust Adversarial Concept Erasure in Diffusion Models
• Minimizing Targeted Activations: Input-Only Suppression of Evaluation-Awareness Latents in Large Language Models
• TokenAR: Multiple Subject Generation via Autoregressive Token-level enhancement
• AutoPrompt: Automated Red-Teaming of Text-to-Image Models via LLM-Driven Adversarial Prompts
• Seeing Through the Brain: New Insights from Decoding Visual Stimuli with fMRI
• All You Need is One: Capsule Prompt Tuning with a Single Vector
• ArenaBencher: Automatic Benchmark Evolution via Multi-Model Competitive Evaluation
• RL Is a Hammer and LLMs Are Nails: A Simple Reinforcement Learning Recipe for Strong Prompt Injection
• Prompt and Parameter Co-Optimization for Large Language Models
• A Survey on Backdoor Threats in Large Language Models (LLMs): Attacks, Defenses, and Evaluation Methods
• No Loss, No Gain: Gated Refinement and Adaptive Compression for Prompt Optimization
• Hierarchical Representation Matching for CLIP-based Class-Incremental Learning
• Semantic Representation Attack against Aligned Large Language Models
• LLM Jailbreak Detection for (Almost) Free!
• Instance-level Randomization: Toward More Stable LLM Evaluations
• Do It Yourself (DIY): Modifying Images for Poems in a Zero-Shot Setting Using Weighted Prompt Manipulation
• NeuroStrike: Neuron-Level Attacks on Aligned LLMs
• Prompt Pirates Need a Map: Stealing Seeds helps Stealing Prompts
• PromptCOS: Towards Content-only System Prompt Copyright Auditing for LLMs
• PromptFlare: Prompt-Generalized Defense via Cross-Attention Decoy in Diffusion-Based Inpainting
• Memory Limitations of Prompt Tuning in Transformers
• Universal and Transferable Adversarial Attack on Large Language Models Using Exponentiated Gradient Descent
• Towards Effective Prompt Stealing Attack against Text-to-Image Diffusion Models
• Multi-module GRPO: Composing Policy Gradients and Prompt Optimization for Language Model Programs

Discussions

• paper on prompt tuning instead of writing them by hand, i haven't read in on this in a while so it might be better now arxiv.org/abs/2302.03668 [bsky, 16 points, 0 comments]

Related

• Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models
• IPGO: Indirect Prompt Gradient Optimization for Parameter-Efficient Prompt-level Fine-Tuning on Text-to-Image Models
• Optimizing Prompts for Text-to-Image Generation
• Manipulating Embeddings of Stable Diffusion Prompts
• Batch-Instructed Gradient for Prompt Evolution:Systematic Prompt Optimization for Enhanced Text-to-Image Synthesis