2026
CLM-LR-DRAFT-ATTENTION-1<= O(n²) score interactions; reduced intermediate memory with IO-aware tilingASYMPTOTIC BOUNDFlashAttention: Fast and Memory-Efficient Exact Attention with IO-AwarenessEvidence: 1 linked source records
SOURCE CONFIRMEDFor sequence length n and hidden width d, specify peak memory and data movement for exact attention; distinguish quadratic score-matrix materialization from IO-aware kernels such as FlashAttention.
Current frontiers derived from accepted Claims.
Assertions tied to evidence, attribution, and review.
The frontier as it changed over time.
Only accepted Claims matching the current specification contribute to the displayed bounds. Strict inequalities remain open; contradictory Claims require editorial review.