对比双模型与单模型推理，发现后者在相同资源下表现更优。
Exploring System 1 and 2 communication for latent reasoning in LLMs
- 用双模块交换潜在信息，测试通信效率提升方法
- 联合微调优于增加通道容量，但整体提升有限
- 单模型共享表示效果接近双模型，适合实际部署
大语言模型的推理应独立于主模型，还是嵌入同一前向传播中？本文研究双架构潜在推理：一个流畅的基模型与协处理器交换潜在消息，并验证两个假设以改进刘等（2024）的方法：（H1）提高信道容量；（H2）通过联合微调学习通信。在 GPT-2 与 Qwen-3 上匹配潜变量令牌预算条件下，H2 始终最优，而 H1 仅带来小幅提升。统一的软嵌入基线——即单模型使用相同前向过程与共享表征，且拥有相同比特预算——几乎追平 H2 并超越 H1，表明当前双模型设计更多是增加计算量而非质的改进。在 GSM8K、ProsQA 及递归分支因子递增的倒计时压力测试中，扩大潜变量预算至小值以上未能提升鲁棒性。潜在空间分析显示子空间重叠严重，专业化不足，与推理能力弱增长一致。结论：双模型潜在推理在理论上仍具潜力，但需明确目标与训练策略以塑造用于算法规划的潜在空间。
原文摘要 · Abstract (English)
Should LLM reasoning live in a separate module, or within a single model's forward pass and representational space? We study dual-architecture latent reasoning, where a fluent Base exchanges latent messages with a Coprocessor, and test two hypotheses aimed at improving latent communication over Liu et al. (2024): (H1) increase channel capacity; (H2) learn communication via joint finetuning. Under matched latent-token budgets on GPT-2 and Qwen-3, H2 is consistently strongest while H1 yields modest gains. A unified soft-embedding baseline, a single model with the same forward pass and shared representations, using the same latent-token budget, nearly matches H2 and surpasses H1, suggesting current dual designs mostly add compute rather than qualitatively improving reasoning. Across GSM8K, ProsQA, and a Countdown stress test with increasing branching factor, scaling the latent-token budget beyond small values fails to improve robustness. Latent analyses show overlapping subspaces with limited specialization, consistent with weak reasoning gains. We conclude dual-model latent reasoning remains promising in principle, but likely requires objectives and training schedules that explicitly shape latent spaces for algorithmic planning.
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。