Skip to content

vllm.models.beam.nvidia.projection

Packed attention projections using shared low-latency GEMMs.