2023-12-05 - Mamba Linear-Time Sequence Modeling with Selective State Spaces

• transformers quadratic with input length
  • do well modeling interactions in a fixed window size, but can't model beyond the supported window
• mamba
  • linear in sequence length
  • 5x higher throughput than transformers
  • beats large transformers
• Structured State Space Sequence Models (SSMs) (S4)

TODO

• continue