Tagged "inference-speed-optimization"

• Dynamic Expert Cache in llama.cpp Achieves 27% Faster Inference on Large MoE Models 15 April 2026