用单个隐变量压缩推理过程，速度提升10倍且准确率损失仅2%。
OneLatent: Single-Token Compression for Visual Latent Reasoning
- 将推理步骤转为图像生成监督信号，实现单隐变量压缩。
- 平均输出长度缩短11倍，长链推理准确率达97.8%以上。
- 适合需要低延迟、高压缩的部署场景，如边缘计算。
思维链（CoT）提示能提升推理能力，但通常使推理成本增加一到两个数量级。为此，我们提出OneLatent框架，通过渲染文本推理步骤为图像，并利用DeepSeek-OCR隐藏状态进行监督，将中间推理过程压缩为单个隐变量。借助图像生成的确定性监督信号，可直接检查与审计，无需模型输出冗长文本。在多个基准测试中，OneLatent将平均输出长度减少11倍，相比文本式思维链仅损失2.21%的平均准确率，同时输出令牌贡献度提升6.8倍。在长链逻辑推理任务中，其在ProntoQA上达到99.80%准确率，在ProsQA上达97.80%，压缩比最高达87.4倍，支持压缩约束下的泛化能力。
原文摘要 · Abstract (English)
Chain-of-thought (CoT) prompting improves reasoning but often increases inference cost by one to two orders of magnitude. To address these challenges, we present \textbf{OneLatent}, a framework that compresses intermediate reasoning into a single latent token via supervision from rendered CoT images and DeepSeek-OCR hidden states. By rendering textual steps into images, we obtain a deterministic supervision signal that can be inspected and audited without requiring the model to output verbose textual rationales. Across benchmarks, OneLatent reduces average output length by $11\times$ with only a $2.21\%$ average accuracy drop relative to textual CoT, while improving output token contribution (OTC) by $6.8\times$. On long-chain logical reasoning, OneLatent reaches $99.80\%$ on ProntoQA and $97.80\%$ on ProsQA with one latent token, with compression up to $87.4\times$, supporting compression-constrained generalization.
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。