参考链接：

jupyter ai develop 开发文档

https://jupyter-ai.readthedocs.io/en/latest/developers/index.html

langchain custom LLM 开发文档

https://python.langchain.com/v0.1/docs/modules/model_io/llms/custom_llm/

stackoverflow :intergrate Local LLM with jupyter ai question

https://stackoverflow.com/questions/78989389/jupyterai-local-llm-integration/78989646#78989646

作者krassowski blog ,关于jupyter lab 有117个post

https://stackoverflow.com/users/6646912/krassowski 

====================================

思路

1。Briefly, define the CustomLLM with something like:

from typing import Any, Dict, Iterator, List, Mapping, Optional

from langchain_core.callbacks.manager import CallbackManagerForLLMRun
from langchain_core.language_models.llms import LLM
from langchain_core.outputs import GenerationChunk


class CustomLLM(LLM):

    def _call(
        self,
        prompt: str,
        stop: Optional[List[str]] = None,
        run_manager: Optional[CallbackManagerForLLMRun] = None,
        **kwargs: Any,
    ) -> str:
        payload = ... # TODO: pass `prompt` to payload here
        # TODO: define `headers`
        response = requests.request(method="POST", url="10.1xx.1xx.50:8084/generate", headers=headers, data=payload)
        return response.text  # TODO: change it accordingly

    @property
    def _llm_type(self) -> str:
        return "custom"

2。 create MyProvider

# my_package/my_provider.py
from jupyter_ai_magics import BaseProvider


class MyProvider(BaseProvider, CustomLLM):
    id = "my_provider"
    name = "My Provider"
    model_id_key = "model"
    models = [
        "your_model"
    ]
    def __init__(self, **kwargs):
        model_id = kwargs.get("model_id")
        # you can use `model_id` in `CustomLLM` to change models within provider
        super().__init__(**kwargs)

3。define an entrypoint 程序入口，配置pyproject.toml

# my_package/pyproject.toml
[project]
name = "my_package"
version = "0.0.1"

[project.entry-points."jupyter_ai.model_providers"]
my-provider = "my_provider:MyProvider"

=================================

部署

cd mypackage/
pip install -e .