Glossary · Infrastructure

Flash Attention

An optimized algorithm for computing the attention mechanism that reduces memory usage and increases speed by minimizing data movement between GPU memory levels, widely adopted across modern model serving stacks.