| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Example online usage of sequence reward models. |
| 4 | |
| 5 | Run `vllm serve <model> --runner pooling` |
| 6 | to start up the server in vLLM. e.g. |
| 7 | |
| 8 | vllm serve Skywork/Skywork-Reward-V2-Qwen3-0.6B |
| 9 | |
| 10 | The key distinction between sequence classification and token classification |
| 11 | lies in their output granularity: sequence classification produces a single |
| 12 | result for an entire input sequence, whereas token classification yields a |
| 13 | result for each individual token within the sequence. |
| 14 | """ |
| 15 | |
| 16 | import argparse |
| 17 | import pprint |
| 18 | |
| 19 | import requests |
| 20 | |
| 21 | |
| 22 | def post_http_request(prompt: dict, api_url: str) -> requests.Response: |
| 23 | headers = {"User-Agent": "Test Client"} |
| 24 | response = requests.post(api_url, headers=headers, json=prompt) |
| 25 | return response |
| 26 | |
| 27 | |
| 28 | def parse_args(): |
| 29 | parser = argparse.ArgumentParser() |
| 30 | parser.add_argument("--host", type=str, default="localhost") |
| 31 | parser.add_argument("--port", type=int, default=8000) |
| 32 | |
| 33 | return parser.parse_args() |
| 34 | |
| 35 | |
| 36 | def main(args): |
| 37 | base_url = f"http://{args.host}:{args.port}" |
| 38 | models_url = base_url + "/v1/models" |
| 39 | pooing_url = base_url + "/pooling" |
| 40 | |
| 41 | response = requests.get(models_url) |
| 42 | model = response.json()["data"][0]["id"] |
| 43 | |
| 44 | # Input like Completions API |
| 45 | prompt = {"model": model, "input": "vLLM is great!"} |
| 46 | pooling_response = post_http_request(prompt=prompt, api_url=pooing_url) |
| 47 | print("-" * 50) |
| 48 | print("Pooling Response:") |
| 49 | pprint.pprint(pooling_response.json()) |
| 50 | print("-" * 50) |
| 51 | |
| 52 | # Input like Chat API |
| 53 | prompt = { |
| 54 | "model": model, |
| 55 | "messages": [ |
| 56 | { |
| 57 | "role": "user", |
| 58 | "content": [{"type": "text", "text": "vLLM is great!"}], |
| 59 | } |
| 60 | ], |
| 61 | } |
| 62 | pooling_response = post_http_request(prompt=prompt, api_url=pooing_url) |
| 63 | print("Pooling Response:") |
| 64 | pprint.pprint(pooling_response.json()) |
| 65 | print("-" * 50) |
| 66 | |
| 67 | |
| 68 | if __name__ == "__main__": |
| 69 | args = parse_args() |
| 70 | main(args) |