| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | # Adapted from https://huggingface.co/boltuix/NeuroBERT-NER |
| 4 | |
| 5 | """Example online usage of Pooling API for Named Entity Recognition (NER). |
| 6 | |
| 7 | Run `vllm serve <model> --runner pooling` |
| 8 | to start up the server in vLLM. e.g. |
| 9 | |
| 10 | vllm serve boltuix/NeuroBERT-NER |
| 11 | """ |
| 12 | |
| 13 | import argparse |
| 14 | |
| 15 | import requests |
| 16 | import torch |
| 17 | |
| 18 | |
| 19 | def post_http_request(prompt: dict, api_url: str) -> requests.Response: |
| 20 | headers = {"User-Agent": "Test Client"} |
| 21 | response = requests.post(api_url, headers=headers, json=prompt) |
| 22 | return response |
| 23 | |
| 24 | |
| 25 | def parse_args(): |
| 26 | parser = argparse.ArgumentParser() |
| 27 | parser.add_argument("--host", type=str, default="localhost") |
| 28 | parser.add_argument("--port", type=int, default=8000) |
| 29 | parser.add_argument("--model", type=str, default="boltuix/NeuroBERT-NER") |
| 30 | |
| 31 | return parser.parse_args() |
| 32 | |
| 33 | |
| 34 | def main(args): |
| 35 | from transformers import AutoConfig, AutoTokenizer |
| 36 | |
| 37 | api_url = f"http://{args.host}:{args.port}/pooling" |
| 38 | model_name = args.model |
| 39 | |
| 40 | # Load tokenizer and config |
| 41 | tokenizer = AutoTokenizer.from_pretrained(model_name) |
| 42 | config = AutoConfig.from_pretrained(model_name) |
| 43 | label_map = config.id2label |
| 44 | |
| 45 | # Input text |
| 46 | text = "Barack Obama visited Microsoft headquarters in Seattle on January 2025." |
| 47 | prompt = {"model": model_name, "input": text} |
| 48 | |
| 49 | pooling_response = post_http_request(prompt=prompt, api_url=api_url) |
| 50 | |
| 51 | # Run inference |
| 52 | output = pooling_response.json()["data"][0] |
| 53 | logits = torch.tensor(output["data"]) |
| 54 | predictions = logits.argmax(dim=-1) |
| 55 | inputs = tokenizer(text, return_tensors="pt") |
| 56 | |
| 57 | # Map predictions to labels |
| 58 | tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0]) |
| 59 | labels = [label_map[p.item()] for p in predictions] |
| 60 | assert len(tokens) == len(predictions) |
| 61 | |
| 62 | # Print results |
| 63 | for token, label in zip(tokens, labels): |
| 64 | if token not in tokenizer.all_special_tokens: |
| 65 | print(f"{token:15} → {label}") |
| 66 | |
| 67 | |
| 68 | if __name__ == "__main__": |
| 69 | args = parse_args() |
| 70 | main(args) |