Description 📖

This PR aims to allow external python package to contribute to the LlamaChatCompletionHandlerRegistry.

Fixes #1715

Security 🔒

This feature is enabled by default, but we could disable it, the idea is to give some control to the users.

Example 🔍

Package outside llama-cpp-python can declare an entrypoint such as bellow

let's say you have a packaged called timmins-plugin-fancy. You can declare an entrypoint as followed

[project.entry-points."llama_cpp_python.register_chat_format"]
load = "timmins-plugin-fancy.llama_cpp_python:load"

Then when the server would start they would receive a LlamaChatCompletionHandlerRegistry instance, which can be used to register custom chat format handler.

Inside your timmins-plugin-fancy package you could have the following

timmins-plugin-fancy
├── pyproject.toml        # and/or setup.cfg, setup.py
└── llama_cpp_python
    └── timmins_plugin_fancy
        └── __init__.py

Here is what the __init__.py could look likes.

import llama_cpp.llama as llama
from llama_cpp.llama_chat_format import LlamaChatCompletionHandlerRegistry

def load(registry: LlamaChatCompletionHandlerRegistry):
    registry.register_chat_completion_handler('timmins-chat', timmins_new_chat_handler, overwrite=True)