ORCIDGitHubLinkedIn
RSS Curriculum Vitae TR

KV Cache

Turkish equivalent: Anahtar-değer önbelleğiDomain: Large Language Models

A cache of previously computed attention key and value tensors used to avoid recomputing earlier sequence states during autoregressive generation.

  • Context Window
  • Multi-Head Attention
  • Speculative Decoding
  • Quantization

Updated:

Related concepts

Large Language Models