Discrete Prompt Compression with Reinforcement Learning

TL;DR

This paper introduces PCRL, a reinforcement learning-based method for discrete prompt compression that reduces token count in prompts for language models, enhancing efficiency and reusability without requiring gradient access or labeled data.

Contribution

PCRL is a novel discrete prompt compression technique using reinforcement learning, addressing interpretability, fixed token limitations, and black-box API interaction issues.

Findings

Achieves 24.6% average token reduction in prompts.

Maintains performance across various LMs.

Policy transferability to larger models.

Abstract

Compressed prompts aid instruction-tuned language models (LMs) in overcoming context window limitations and reducing computational costs. Existing methods, which primarily based on training embeddings, face various challenges associated with interpretability, the fixed number of embedding tokens, reusability across different LMs, and inapplicability when interacting with black-box APIs. This study proposes prompt compression with reinforcement learning (PCRL), which is a discrete prompt compression method that addresses these issues. The proposed PCRL method utilizes a computationally efficient policy network that edits prompts directly. The training approach employed in the proposed PCRLs can be applied flexibly to various types of LMs, including both decoder-only and encoder-decoder architecture and it can be trained without gradient access to the LMs or labeled data. The proposed PCRL achieves an average reduction of 24.6% in terms of the token count across various instruction prompts while maintaining sufficient performance. In addition, we demonstrate that the learned policy can be transferred to larger LMs, and through a comprehensive analysis, we explore the token importance within the prompts. Our code is accessible at https://github.com/nenomigami/PromptCompressor.