HFLLMAPI / README.md
Husnain
Update README.md
26eb68a unverified
|
raw
history blame
4.27 kB
metadata
title: HF LLM API
emoji: ☪️
colorFrom: gray
colorTo: gray
sdk: docker
app_port: 23333

HF-LLM-API

Huggingface LLM Inference API in OpenAI message format.

Original Project link: https://github.com/Hansimov/HF-LLM-API

Features

  • Available Models (2024/04/20):
    • mistral-7b, mixtral-8x7b, nous-mixtral-8x7b, gemma-7b, command-r-plus, llama3-70b, zephyr-141b, gpt-3.5-turbo
    • Adaptive prompt templates for different models
  • Support OpenAI API format
    • Enable api endpoint via official openai-python package
  • Support both stream and no-stream response
  • Support API Key via both HTTP auth header and env variable
  • Docker deployment

Run API service

Run in Command Line

Install dependencies:

# pipreqs . --force --mode no-pin
pip install -r requirements.txt

Run API:

python -m apis.chat_api

Run via Docker

Docker build:

sudo docker build -t hf-llm-api:1.1.3 . --build-arg http_proxy=$http_proxy --build-arg https_proxy=$https_proxy

Docker run:

# no proxy
sudo docker run -p 23333:23333 hf-llm-api:1.1.3

# with proxy
sudo docker run -p 23333:23333 --env http_proxy="http://<server>:<port>" hf-llm-api:1.1.3

API Usage

Using openai-python

See: examples/chat_with_openai.py

from openai import OpenAI

# If runnning this service with proxy, you might need to unset `http(s)_proxy`.
base_url = "http://127.0.0.1:23333"
# Your own HF_TOKEN
api_key = "hf_xxxxxxxxxxxxxxxx"
# use below as non-auth user
# api_key = "sk-xxx"

client = OpenAI(base_url=base_url, api_key=api_key)
response = client.chat.completions.create(
    model="nous-mixtral-8x7b",
    messages=[
        {
            "role": "user",
            "content": "what is your model",
        }
    ],
    stream=True,
)

for chunk in response:
    if chunk.choices[0].delta.content is not None:
        print(chunk.choices[0].delta.content, end="", flush=True)
    elif chunk.choices[0].finish_reason == "stop":
        print()
    else:
        pass

Using post requests

See: examples/chat_with_post.py

import ast
import httpx
import json
import re

# If runnning this service with proxy, you might need to unset `http(s)_proxy`.
chat_api = "http://127.0.0.1:23333"
# Your own HF_TOKEN
api_key = "hf_xxxxxxxxxxxxxxxx"
# use below as non-auth user
# api_key = "sk-xxx"

requests_headers = {}
requests_payload = {
    "model": "nous-mixtral-8x7b",
    "messages": [
        {
            "role": "user",
            "content": "what is your model",
        }
    ],
    "stream": True,
}

with httpx.stream(
    "POST",
    chat_api + "/chat/completions",
    headers=requests_headers,
    json=requests_payload,
    timeout=httpx.Timeout(connect=20, read=60, write=20, pool=None),
) as response:
    # https://docs.aiohttp.org/en/stable/streams.html
    # https://github.com/openai/openai-cookbook/blob/main/examples/How_to_stream_completions.ipynb
    response_content = ""
    for line in response.iter_lines():
        remove_patterns = [r"^\s*data:\s*", r"^\s*\[DONE\]\s*"]
        for pattern in remove_patterns:
            line = re.sub(pattern, "", line).strip()

        if line:
            try:
                line_data = json.loads(line)
            except Exception as e:
                try:
                    line_data = ast.literal_eval(line)
                except:
                    print(f"Error: {line}")
                    raise e
            # print(f"line: {line_data}")
            delta_data = line_data["choices"][0]["delta"]
            finish_reason = line_data["choices"][0]["finish_reason"]
            if "role" in delta_data:
                role = delta_data["role"]
            if "content" in delta_data:
                delta_content = delta_data["content"]
                response_content += delta_content
                print(delta_content, end="", flush=True)
            if finish_reason == "stop":
                print()