研究人员开发了新方法来提高大型语言模型中链式思考（CoT）推理的效率。HybridThinker引入了一种训练方案，在保留详细思考步骤和将其压缩到内存令牌之间取得平衡，以相似的推理时间实现了最先进的准确性。HMPO提供了一个成本效益高、单阶段的强化学习框架，能够自适应地压缩CoT，在各种任务和模型规模上显著减少令牌数量，同时准确性损失可忽略不计。另一项研究探讨了CoT和循环Transformer的内存机制，强调了与完整序列状态循环或CoT暂存器不同，压缩循环受其循环状态大小的限制。 AI
影响 这些在CoT压缩和内存管理方面的进步可能带来更强大、更高效的LLM，以应对复杂的推理任务。
排序理由 多篇研究论文介绍了提高LLM推理效率的新颖技术。
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
arXiv:2606.03768v1 Announce Type: new Abstract: Extended chain-of-thought (CoT) traces improve LLM reasoning but incur substantial computational and memory costs. While existing CoT compression methods mitigate this by condensing thought steps into compact representations via mem…
Extended chain-of-thought (CoT) traces improve LLM reasoning but incur substantial computational and memory costs. While existing CoT compression methods mitigate this by condensing thought steps into compact representations via memory tokens and retaining only these representati…
arXiv:2606.01934v1 Announce Type: cross Abstract: Large language models achieve remarkable performance via extended chain-of-thought (CoT) reasoning, yet this lengthy process incurs substantial inference overhead. Existing CoT compression methods struggle with inflexible manual l…
Large language models achieve remarkable performance via extended chain-of-thought (CoT) reasoning, yet this lengthy process incurs substantial inference overhead. Existing CoT compression methods struggle with inflexible manual length budgets, computationally expensive multi-sta…
arXiv:2605.30757v1 Announce Type: new Abstract: Chain-of-thought prompting and looped Transformers both give a fixed model more test-time computation, but they differ in what they remember. Chain-of-thought stores intermediate state in generated tokens that remain in the context,…