Add support for gguf (original) (raw)

The latest llama.cpp development has deprecated the ggml format in favor of a new gguf format.

llama.cpp has chosen to break their API and make ggml models useless. The goal for llama.clj is to upgrade without breaking backwards compatibility. More research is required, but the initial plan is something like:

• treat the new raw API for llama.cpp as a separate library
• create a protocol for any shared functionality and implement it for both the ggml version and the latest version
• create independent builds that can be included independently or together
• Add better support for including your own llama.cpp build