v0.18.27: Is the PR faster?

A llama.cpp PR build in Build / Update now has a Compare speed button. Pick a model and LlamaForge runs llama-bench (pp512 / tg128, 3 runs each) on it with your build, then with the PR build: same model file, same GPU settings, every other model unloaded meanwhile and loaded back afterwards.

The card shows both speeds with llama-bench's own spread and the change in percent. A change smaller than the two spreads added together is marked within noise, so jitter doesn't pass for a speedup. Copy as markdown puts the table on your clipboard for the PR's thread.

See Build & Update.