EPISODE · Mar 14, 2025 · 8 MIN
AI前沿:从文本分块到知识模块的创新突破
from AI可可AI生活
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在小宇宙查看该单集文稿
Embed this episode
Ready to play
AI前沿:从文本分块到知识模块的创新突破
No transcript for this episode yet
Similar Episodes
No similar episodes found.