| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Example Python client for OpenAI Chat Completion using vLLM API server |
| 4 | NOTE: start a supported chat completion model server with `vllm serve`, e.g. |
| 5 | vllm serve meta-llama/Llama-2-7b-chat-hf |
| 6 | """ |
| 7 | |
| 8 | import argparse |
| 9 | |
| 10 | from openai import OpenAI |
| 11 | |
| 12 | # Modify OpenAI's API key and API base to use vLLM's API server. |
| 13 | openai_api_key = "EMPTY" |
| 14 | openai_api_base = "http://localhost:8000/v1" |
| 15 | |
| 16 | messages = [ |
| 17 | {"role": "system", "content": "You are a helpful assistant."}, |
| 18 | {"role": "user", "content": "Who won the world series in 2020?"}, |
| 19 | { |
| 20 | "role": "assistant", |
| 21 | "content": "The Los Angeles Dodgers won the World Series in 2020.", |
| 22 | }, |
| 23 | {"role": "user", "content": "Where was it played?"}, |
| 24 | ] |
| 25 | |
| 26 | |
| 27 | def parse_args(): |
| 28 | parser = argparse.ArgumentParser(description="Client for vLLM API server") |
| 29 | parser.add_argument( |
| 30 | "--stream", action="store_true", help="Enable streaming response" |
| 31 | ) |
| 32 | return parser.parse_args() |
| 33 | |
| 34 | |
| 35 | def main(args): |
| 36 | client = OpenAI( |
| 37 | # defaults to os.environ.get("OPENAI_API_KEY") |
| 38 | api_key=openai_api_key, |
| 39 | base_url=openai_api_base, |
| 40 | ) |
| 41 | |
| 42 | models = client.models.list() |
| 43 | model = models.data[0].id |
| 44 | |
| 45 | # Chat Completion API |
| 46 | chat_completion = client.chat.completions.create( |
| 47 | messages=messages, |
| 48 | model=model, |
| 49 | stream=args.stream, |
| 50 | ) |
| 51 | |
| 52 | print("-" * 50) |
| 53 | print("Chat completion results:") |
| 54 | if args.stream: |
| 55 | for c in chat_completion: |
| 56 | print(c) |
| 57 | else: |
| 58 | print(chat_completion) |
| 59 | print("-" * 50) |
| 60 | |
| 61 | |
| 62 | if __name__ == "__main__": |
| 63 | args = parse_args() |
| 64 | main(args) |