Slim attention: cut your context memory in half without loss -- K-cache is all you need for MHA

Documenti analoghi

• KV-weights are all you need for skipless transformers
  di: Graef, Nils
  Pubblicazione: (2024)
• FlashNorm: Fast Normalization for Transformers
  di: Graef, Nils, et al.
  Pubblicazione: (2024)
• Linear attention is (maybe) all you need (to understand transformer optimization)
  di: Ahn, Kwangjun, et al.
  Pubblicazione: (2023)
• Transformer tricks: Precomputing the first layer
  di: Graef, Nils
  Pubblicazione: (2024)
• Is attention all you need in medical image analysis? A review
  di: Papanastasiou, Giorgos, et al.
  Pubblicazione: (2023)