Less is More: DocString Compression in Code Generation

2026-04-11

AI Summary

To address the problem of excessive prompt length and high inference cost in code generation caused by redundant DocStrings, this paper proposes ShortenDoc, a task-specific DocString compression method. Unlike generic compression techniques—which achieve only ~10% compression—ShortenDoc is the first approach to jointly leverage code functionality intent recognition and structure-aware strategies. It employs an iterative pruning-and-rewriting mechanism constrained by semantic fidelity, achieving high compression rates of 25–40% without degrading code generation quality. Extensive experiments across six code generation benchmarks, five open-source LLMs (1B–10B parameters), and GPT-4o demonstrate that ShortenDoc significantly reduces token overhead while preserving code accuracy at baseline levels. This breaks the performance bottleneck of existing methods, establishing a new state-of-the-art for efficient DocString-aware code generation.

Technology Category

Natural Language Processing: Code Generation / Program Synthesis from Natural LanguageData Mining & Knowledge Management: Data CompressionMachine Learning: Learning on the Edge & Model Compression

Application Category

Economics, Online Markets and Human Computation: Cost models of using LLMs in production systemsSearch and Retrieval-Augmented AI: Web evaluation methodologies and metricsSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

The widespread use of Large Language Models (LLMs) in software engineering has intensified the need for improved model and resource efficiency. In particular, for neural code generation, LLMs are used to translate function/method signature and DocString to executable code. DocStrings which capture user re quirements for the code and used as the prompt for LLMs, often contains redundant information. Recent advancements in prompt compression have shown promising results in Natural Language Processing (NLP), but their applicability to code generation remains uncertain. Our empirical study show that the state-of-the-art prompt compression methods achieve only about 10% reduction, as further reductions would cause significant performance degradation. In our study, we propose a novel compression method, ShortenDoc, dedicated to DocString compression for code generation. Our extensive experiments on six code generation datasets, five open-source LLMs (1B to 10B parameters), and one closed-source LLM GPT-4o confirm that ShortenDoc achieves 25-40% compression while preserving the quality of generated code, outperforming other baseline methods at similar compression levels. The benefit of this research is to improve efficiency and reduce the cost while maintaining the quality of the generated code, especially when calling third-party APIs, and is able to reduce the token processing cost by 25-40%.

Problem

• Compress DocStrings for efficient code generation
• Reduce token costs without degrading code quality
• Optimize LLM prompts by removing redundant information

Innovation

• DocString compression for code generation
• 25-40% compression with quality preservation
• Reduces token processing cost significantly