TROPT: An Open Framework for Unifying and Advancing Discrete Text Optimization

2026-06-22

AI Summary

Existing discrete text optimizers lack a unified framework, hindering reusability and systematic comparison across diverse models, objectives, and tasks. This work proposes the first modular open-source framework that standardizes interfaces to integrate over 30 optimization algorithms—including both white-box and black-box methods—and more than 15 loss functions. The framework enables flexible composition of models, objectives, and optimization strategies within an end-to-end customizable pipeline. By significantly lowering the barrier to entry for both application and innovation, it has been successfully deployed in large-scale jailbreaking strategy comparisons and effectively transferred to novel tasks such as corpus poisoning, demonstrating its broad applicability and practical utility.

Abstract

Discrete text-trigger optimization -- searching for text sequences that, when ingested by a model, steer it toward a specified objective -- underpins model red-teaming (e.g., LLM jailbreaks), as well as auditing and interpretability. However, the current state of discrete optimizers hinders their adoption and progress. First, existing optimizers, when open-sourced at all, are scattered across research codebases tied to specific models, objectives, and problem domains. Second, optimizer variants proliferate, each requiring engineering overhead to use or extend, and remaining hard to compare head-to-head. Together, these raise the bar for adopting optimizers in existing or new domains, and for advancing them via new strategies. We address these gaps with TROPT, the first open-source framework that unifies discrete optimizers' execution and standardizes their development under a single interface. TROPT makes it easy to customize end-to-end optimization recipes by swapping any component -- models, objectives, and optimizers -- extending its reach across domains and new applications. TROPT currently ships with 30+ optimization recipes -- covering applications such as jailbreaking and probing model internals -- built from 15+ optimizers (spanning white-box to black-box access) and 15+ losses, from foundational to state-of-the-art methods. Demonstrating its utility, we leverage TROPT in several studies: (i) controlled, large-scale experiments comparing and enhancing optimization strategies for LLM jailbreaks, revealing potent-yet-underadopted techniques; and (ii) porting optimizers from one domain (e.g., LLM jailbreak) to new domains (e.g., corpus-poisoning embedding model). In all, TROPT significantly lowers the barrier to adopting and advancing discrete text optimization.

Problem

• discrete text optimization
• optimizer standardization
• model red-teaming
• LLM jailbreak
• open framework

Innovation

• discrete text optimization
• unified framework
• TROPT
• optimizer standardization
• LLM red-teaming