HKUSTDial/flash-sparse-attention
**Brief Summary (2‑3 sentences)** HKUSTDial’s *flash‑sparse‑attention* library provides a trainable, fast, and memory‑efficient implementation of sparse attenti
Original source**Brief Summary (2‑3 sentences)** HKUSTDial’s *flash‑sparse‑attention* library provides a trainable, fast, and memory‑efficient implementation of sparse attenti
Original source