| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Example online usage of Score API. |
| 4 | |
| 5 | Run `vllm serve <model> --runner pooling` to start up the server in vLLM. |
| 6 | """ |
| 7 | |
| 8 | import argparse |
| 9 | import pprint |
| 10 | |
| 11 | import requests |
| 12 | |
| 13 | |
| 14 | def post_http_request(prompt: dict, api_url: str) -> requests.Response: |
| 15 | headers = {"User-Agent": "Test Client"} |
| 16 | response = requests.post(api_url, headers=headers, json=prompt) |
| 17 | return response |
| 18 | |
| 19 | |
| 20 | def parse_args(): |
| 21 | parser = argparse.ArgumentParser() |
| 22 | parser.add_argument("--host", type=str, default="localhost") |
| 23 | parser.add_argument("--port", type=int, default=8000) |
| 24 | parser.add_argument("--model", type=str, default="BAAI/bge-reranker-v2-m3") |
| 25 | return parser.parse_args() |
| 26 | |
| 27 | |
| 28 | def main(args): |
| 29 | api_url = f"http://{args.host}:{args.port}/score" |
| 30 | model_name = args.model |
| 31 | |
| 32 | queries = "What is the capital of Brazil?" |
| 33 | documents = "The capital of Brazil is Brasilia." |
| 34 | prompt = {"model": model_name, "queries": queries, "documents": documents} |
| 35 | score_response = post_http_request(prompt=prompt, api_url=api_url) |
| 36 | print("\nPrompt when queries and documents are both strings:") |
| 37 | pprint.pprint(prompt) |
| 38 | print("\nScore Response:") |
| 39 | pprint.pprint(score_response.json()) |
| 40 | |
| 41 | queries = "What is the capital of France?" |
| 42 | documents = [ |
| 43 | "The capital of Brazil is Brasilia.", |
| 44 | "The capital of France is Paris.", |
| 45 | ] |
| 46 | prompt = {"model": model_name, "queries": queries, "documents": documents} |
| 47 | score_response = post_http_request(prompt=prompt, api_url=api_url) |
| 48 | print("\nPrompt when queries is string and documents is a list:") |
| 49 | pprint.pprint(prompt) |
| 50 | print("\nScore Response:") |
| 51 | pprint.pprint(score_response.json()) |
| 52 | |
| 53 | queries = ["What is the capital of Brazil?", "What is the capital of France?"] |
| 54 | documents = [ |
| 55 | "The capital of Brazil is Brasilia.", |
| 56 | "The capital of France is Paris.", |
| 57 | ] |
| 58 | prompt = {"model": model_name, "queries": queries, "documents": documents} |
| 59 | score_response = post_http_request(prompt=prompt, api_url=api_url) |
| 60 | print("\nPrompt when queries and documents are both lists:") |
| 61 | pprint.pprint(prompt) |
| 62 | print("\nScore Response:") |
| 63 | pprint.pprint(score_response.json()) |
| 64 | |
| 65 | |
| 66 | if __name__ == "__main__": |
| 67 | args = parse_args() |
| 68 | main(args) |