Comment by asattarmd

What's the fundamental limitation to context size here? Why can't a model be fine-tuned per codebase, taking the entire code into context (and be continuously trained as it's updated)?

Forgive my naivety, I don't now anything about LLMs.