本期播客精华汇总 MoC: Mixtures of Text Chunking Learners for Retrieval-Augmented Generation System：提出混合分块器MoC，通过动态选择分块策略和生成规则，提升了文本分块质量，显著增强了检索系统的问答能力。 LLMs Know What to Drop: Self-Attention Guided KV Cache Eviction for Efficient Long-Context Inference：推出SAGE-KV方法，用注意力分数指导内存压缩，让AI在处理长文本时更高效，内存节省高达4倍。 Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models：提出块扩散模型BD3-LM，结合两种生成方式，实现快速、灵活的文本生成，质量逼近主流模型。 Plan-and-Act: Improving Planning of Agents for Long-Horizon Tasks：设计Plan-and-Act框架，分开规划与执行并动态调整，让AI在复杂任务中成功率大增。 Training Plug-n-Play Knowledge Modules with Deep Context Distillation：推出知识模块KM和深度蒸馏训练法，为AI提供高效知识注入方案，适合快速学习新文档。完整推介：https://mp.weixin.qq.com/s/CSB7asQUDcwSlUDJxwIBLg

Ask this episode

“What did the council decide on item 42?” — answered with the speaker and the moment.

Transcript

Related