BWhat changed
Combined asynchronous data movement, overlapped matrix and softmax operations, and low-precision processing to accelerate attention on Hopper GPUs.
Scientific scope
Combined asynchronous data movement, overlapped matrix and softmax operations, and low-precision processing to accelerate attention on Hopper GPUs. FP8 introduces numerical approximation; reported accuracy comparisons do not imply bitwise identity or universal precision gains. The hardware-specific algorithms extend, rather than invent, memory-aware attention.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. FP8 introduces numerical approximation; reported accuracy comparisons do not imply bitwise identity or universal precision gains. The hardware-specific algorithms extend, rather than invent, memory-aware attention.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision
- Tri Dao: FlashAttention-3 research account
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction