llama.cpp performance optimizations

A PR in llama.cpp removes redundant softmax+sort in Top-N-Sigma sampler, boosting inference speed by 50% on Gemma-4-E4B-Q8_0.

One continuously updated timeline instead of dozens of separate articles. New developments are appended as the story evolves.

1. UpdateJun 22, 2026, 05:18 PM 71%

   llama.cpp PR #22645 optimizes Top-N-Sigma sampler, boosting inference speed by 50%

   A PR in llama.cpp removes redundant softmax+sort in Top-N-Sigma sampler, boosting inference speed by 50% on Gemma-4-E4B-Q8_0.

   Read the full story →