Fig.1

Concept

DSA

DSA (DeepSeek Sparse Attention) is an attention mechanism that avoids computing the full quadratic attention matrix by having each query attend to only a selected subset of key/value tokens. Standard self-attention costs O(n^2) in sequence length, so at 256K tokens the…

The rest of “DSA” is a premium feature: every concept in the library gets a precise, practitioner-focused write-up like this one, cross-linked straight from the paper summaries that use it.

Log in to unlock

← Back to the library