Flash Attention

Speed up long-sequence transformer training and inference.

By Orchestra Research · Official Docs · Date unknown

Flash Attention provides 2-4x speedup and 10-20x memory reduction for transformer attention through IO-aware tiling and recomputation.

Platforms: linux, macos.

hermes skills install official/mlops/flash-attention

• #skills
• #optional
• #mlops
• #optimization
• #flash attention
• #attention optimization
• #memory efficiency