压缩思维链冗余 tokens，提升推理效率与准确率
Not All Tokens Are What You Need In Thinking
- 按重要性动态筛选思维链关键 token，实现灵活压缩
- GPQA 上准确率提升 9.1%，推理 token 减少 13.2%
- 适合追求高效推理的模型优化与部署场景
现代推理模型如 OpenAI o1 和 DeepSeek-R1 虽具备强大解题能力，但存在推理延迟高、资源消耗大及过度思考等问题，表现为冗长的思维链（CoT）中充斥大量低效 token。为此，我们提出条件性令牌选择（CTS），一种基于令牌级压缩的框架，通过条件重要性评分识别并保留思维链中最关键的令牌，再用压缩后的思维链训练模型。大量实验表明，CTS 能有效压缩长思维链同时保持强推理性能。在 GPQA 基准上，使用 CTS 训练的 Qwen2.5-14B-Instruct 模型准确率提升 9.1%，推理令牌减少 13.2%（训练令牌减少 13%）。进一步将训练令牌减少 42%，仅导致 5% 准确率下降，但推理令牌减少达 75.8%，凸显现有思维链中冗余严重。
原文摘要 · Abstract (English)
Modern reasoning models, such as OpenAI's o1 and DeepSeek-R1, exhibit impressive problem-solving capabilities but suffer from critical inefficiencies: high inference latency, excessive computational resource consumption, and a tendency toward overthinking -- generating verbose chains of thought (CoT) laden with redundant tokens that contribute minimally to the final answer. To address these issues, we propose Conditional Token Selection (CTS), a token-level compression framework with a flexible and variable compression ratio that identifies and preserves only the most essential tokens in CoT. CTS evaluates each token's contribution to deriving correct answers using conditional importance scoring, then trains models on compressed CoT. Extensive experiments demonstrate that CTS effectively compresses long CoT while maintaining strong reasoning performance. Notably, on the GPQA benchmark, Qwen2.5-14B-Instruct trained with CTS achieves a 9.1% accuracy improvement with 13.2% fewer reasoning tokens (13% training token reduction). Further reducing training tokens by 42% incurs only a marginal 5% accuracy drop while yielding a 75.8% reduction in reasoning tokens, highlighting the prevalence of redundancy in existing CoT.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。