This looks very interesting. Can this be trained on the user's codebase, or is t... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		amelius 11 months ago \| parent \| context \| favorite \| on: Llama.vim – Local LLM-assisted text completion This looks very interesting. Can this be trained on the user's codebase, or is the idea that everything must fit inside the context buffer?

larodi 11 months ago [–]

Models don’t get trained on your codebase in scenarios like Continue and Copilot assisted completion. It’s RAG at best and it depends on how the RAG is built.

So all depends on how it fits in context and it will for the foreseeable time be like it as training is absolutely expensive compared to RAG.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact