Mamba: Linear-Time Sequence Modeling with Selective State Spaces

  • 类型:arxiv
  • 标识:2312.00752
  • 链接:https://arxiv.org/abs/2312.00752
  • 主题:llm-infra
  • 主分类:llm-infra
  • 形态:method
  • 被引:8699
  • 被引来源:Semantic Scholar
  • S2被引:8699
  • OpenAlex被引:1034
  • 影响力被引:1260
  • TLDR:This work identifies that a key weakness of subquadratic-time models based on Transformer architecture is their inability to perform content-based reasoning, and integrates selective SSMs into a simplified end-to-end neural network architecture without attention or even MLP blocks (Mamba).
  • OpenAlex ID:W4389326242
  • OpenAlex DOI:10.48550/arxiv.2312.00752
  • DOI:10.48550/arxiv.2312.00752
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2312.00752
  • OpenAlex更新:2026-08-08
  • 待LLM分类:否
  • 成熟度:research
  • 场景:序列建模、状态空间、高效架构
  • 标题中文:Mamba:基于选择性状态空间的线性时间序列建模
  • TLDR中文:本文指出基于 Transformer 的次二次时间模型的关键缺陷在于无法执行基于内容的推理,并将选择性 SSM 集成到不包含注意力乃至 MLP 块的简化端到端神经网络架构(Mamba)中。
  • 来源文件
  • [OpenAlex discover]
  • [S2 enrich]