Expand description
Attention evaluator for the optimizer loop.
Structs§
- Attention
OptResult - Result of optimizing Attention for a single problem size.
- Attention
Problem Config - A single Attention problem size with its FlashAttention v2 baseline time.
- Real
Attention Evaluator - Evaluates Attention kernel parameters by modeling execution time based on actual kernel characteristics and computing efficiency vs FlashAttention v2.
Functions§
- generate_
attention_ milestone_ json - generate_
attention_ milestone_ report - optimize_
all_ attention_ problems - optimize_
attention_ problem - standard_
attention_ problems - Standard Attention problem configurations with FlashAttention v2 baselines.