llama-patch-changes

To manage the llama.cpp patch queue, you maintain durable patches under third_party/llama.cpp/patches and update build scripts that consume the patched llama.cpp source to ensure changes remain focused and reviewable.

Updating the upstream llama.cpp pin requires modifying the LLAMA_CPP_SHA and the third_party/llama.cpp/upstream.txt file to prevent drift between your local patches and upstream changes.

You can refresh patches from a llama.cpp checkout by regenerating and validating the patch queue while avoiding submodules, keeping all changes focused and reviewable under third_party/llama.cpp/patches.

Patch queue drift occurs when local patches under third_party/llama.cpp/patches are not synchronized with the pinned upstream SHA, requiring you to run patch regeneration and validation workflows for clean checkouts.

Yes, you can alter build scripts that consume patched llama.cpp alongside patch editing, provided you maintain durable patches under third_party/llama.cpp/patches and keep changes focused and reviewable.