摘要：大语言模型（LLM）的最新进展使得提示词变得越来越庞大且复杂。思维链推理（Chain-of-thought reasoning, Wei et al., 2022）以及上下文学习（In-context learning, Brown et al., 2020）等技术频繁导致实际场景中的提示词超过数千 Token，从而增加了推理成本和延迟。诸如 LLMLingua（Jiang et al., 2023）和 Selective Context（Li et al., 2023）等基于模型学习的压缩方法虽然能够实现较高的压缩比，但通常需要辅助语言模型，且存在非确定性（Non-deterministic）。我们提出了一个互补性的问题：在输出质量发生显著退化之前，一套完全基于经典词法 NLP、无需训练、完全确定性且仅运行在 CPU 上的流水线究竟能将压缩推进到何种程度？我们构建了一个由 11 项可独立开关的词法转换组成的可配置流水线——包括停用词移除、填充短语删除、缩略词与简写替换、基于词性的修剪、词形还原、基于 WordNet 的同义词简短替换以及命名实体保留。我们在来自 6 个数据源（Dolly-15k、LMSYS-Chat-1M、WildChat-1M、MMLU、GSM8K、HellaSwag）的 1,242 个纯英文提示词上评估了 15 种配置，覆盖 11 个自动归纳的任务类别，共生成了 18,630 组配对的 GPT-4o-mini 输出。输出保真度通过 BLEU、ROUGE-1/2/L、BERTScore-F1 和 SentenceBERT 余弦相似度进行评估。实验结果显示，最激进的配置实现了 40.3% 的平均 Token 缩减量（标准差 sigma = 9.2），与原始提示词输出相比 BERTScore-F1 达到 0.876；仅使用停用词移除的配置即可实现 29.6% 的 Token 缩减，BERTScore-F1 达到 0.913。本研究针对各个任务类别分别绘制了“压缩率-保真度”帕累托前沿（Pareto frontier），并发现常识推理是激进压缩下的系统性失效场景。为了确保研究的可复现性，全部代码、提示词以及各单元测试结果均已开源。

Abstract:Recent advances in large language models (LLMs) have made prompts increasingly large and complex. Techniques such as chain-of-thought reasoning (Wei et al., 2022) and in-context learning (Brown et al., 2020) frequently push real-world prompts past several thousand tokens, increasing inference cost and latency. Learned compression methods such as LLMLingua (Jiang et al., 2023) and Selective Context (Li et al., 2023) achieve high compression ratios but require auxiliary language models and are non-deterministic. We ask a complementary question: how far can a training-free, fully deterministic, CPU-only pipeline based on classical lexical NLP be pushed before output quality degrades significantly? Eleven toggleable lexical transformations - stopword removal, filler-phrase deletion, contraction and abbreviation substitution, part-of-speech-based pruning, lemmatization, WordNet-driven synonym shortening, and named-entity preservation - are assembled into a configurable pipeline. Fifteen configurations are evaluated on 1,242 English-only prompts from six sources (Dolly-15k, LMSYS-Chat-1M, WildChat-1M, MMLU, GSM8K, HellaSwag), spanning eleven automatically derived task categories, yielding 18,630 paired GPT-4o-mini completions. Output preservation is measured using BLEU, ROUGE-1/2/L, BERTScore-F1, and SentenceBERT cosine similarity. The most aggressive configuration achieves a mean token reduction of 40.3% (sigma = 9.2) at a BERTScore-F1 of 0.876 against the original-prompt output; a stopword-only configuration achieves 29.6% reduction at 0.913. The compression-versus-fidelity Pareto frontier is characterized per task category, with commonsense reasoning a systematic failure mode under aggressive compression. All code, prompts, and per-cell results are released for reproducibility.