Across
- 3. Unaffected by FlashAttention
- 4. FlashAttention's key feature
- 6. Reduced memory usage
- 8. Mechanism in LLMs
- 9. Co-design for future
- 11. Real-world impact of FA-2
- 12. Enabled by longer context
- 13. FlashAttention-2's improvement
Down
- 1. Scaling of attention
- 2. Faster attention mechanism
- 5. Limited resource in LLMs
- 7. Longer sequences enabled
- 10. Application in drug discovery
- 14. Computation across sequence
