Building off an initial request for comments (RFC) patch series posted during the winter holidays, an updated RFC patch series was posted this weekend for LLMinus. LLMinus is an effort led by NVIDIA Linux kernel engineer Sasha Levin to provide a large language model (LLM) assisted merge conflict resolution tool focused on Linux kernel development.
LLMinus aims to aide kernel developers with resolving Git merge conflicts within the Linux kernel. LLMinus has a searchable database of historical conflict resolutions based on the Linux kernel’s Git history. LLMinus relies on semantic embeddings to find similar past conflicts in Git merges to the kernel to help resolve current conflicts in cooperation with LLMs. LLMinus isn’t tied to a single particular LLM but is said to work with any model supporting standard input and ideally the LLMs handling dynamic investigation.
The updated LLMinus patches this weekend add a “–max-tokens” argument with a 100K token default to help prevent prompt overflow errors. There is also semantic conflict detection via build test integration.
Those wanting to learn more about this AI/LLVM-assisted merge conflict resolution project can see the latest patches on the Linux kernel mailing list.
Sasha Levin of NVIDIA has also been using AI/LLMs to help determine kernel patches for back-porting to the Linux stable kernel series he co-maintains.
