svllm 0.3.2

serve your llm

svllm

Create a OpenAI-Compatible API Server for your LLM.

Usage

Install

$ pip install svllm

Prepare

svllm provided following HTTP APIs:

• chat: /v1/chat/completions
• complete: /v1/completions
• embed: /v1/embeddings
• models: /v1/models

You only need to implement the chat, complete, embed function (or part of them).

Set __model__, __owner__ for these functions can affect /v1/models API.

See src/protocol.py for more information.

Then you can use svllm in the following two ways:

CLI

If you have a demo.py file that implements a chat function, you can start it like this:

$ svllm --chat demo:chat
$ svllm-cli # client

Run svllm --help and svllm-cli --help for more information.

Library

You can also use svllm as a library. (also see demo.py)

import uvicorn
import svllm

if __name__ == '__main__':
    svllm.set_chat(svllm.examples.chat)
    svllm.set_complete(svllm.examples.complete)
    svllm.set_embed(svllm.examples.embed)

    app = svllm.create_app()
    uvicorn.run(app, host='0.0.0.0', port=5261)