llama.cpp b11080 → b11211, CUDA 13.4, ROCm 10 (TheRock), sccache on every Windows Ninja job - #456

Merged

bernardladenthin merged 10 commits into

mainfrom

claude/busy-archimedes-lo9uyy

Sep 27, 2026

Commits on Sep 27, 2026

Upgrade llama.cpp from b11080 to b11103
claude
committed
d52e6ce
Upgrade llama.cpp from b11103 to b11104: multi-address --host
claude
committed
7dfb218
Upgrade llama.cpp from b11104 to b11160
claude
committed
a243ed6
Upgrade llama.cpp from b11160 to b11163: llama_batch_ext
claude
committed
2f7fbb8
Upgrade llama.cpp from b11163 to b11209
claude
committed
49932e3
Build the ROCm classifiers against ROCm 10 from TheRock
claude
committed
9907dad
CUDA 13.4, gfx900/gfx906 kept, free-disk-space on heavy Linux jobs
claude
committed
d9c21cd
ROCm: build every GPU target TheRock supports per OS
claude
committed
ea29df7
sccache on every Windows Ninja job, with an uncached retry in build.bat
claude
committed
607b423
Upgrade llama.cpp from b11209 to b11211
claude
committed
1a969ca