Crossword

12345678910111213
Across
  1. 5. Hardware used by FlashAttention
  2. 6. Model using attention mechanism
  3. 7. Multiplication in attention
  4. 8. Mechanism improved by FlashAttention
  5. 9. Length impacting attention speed
  6. 11. Length increased by FlashAttention
  7. 12. Performance gain of FlashAttention
Down
  1. 1. Units of text in long sequences
  2. 2. Resource constraint in attention
  3. 3. Scaling of standard attention
  4. 4. Software co-design focus
  5. 10. Goal of FlashAttention improvements
  6. 13. Processing in FlashAttention-2