[New Model] Support BertForTokenClassification / Named Entity Recognition (NER) task (#24872)

Signed-off-by: wang.yuqi <noooop@126.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>

examples/offline_inference/pooling/README.md

@@ -26,8 +26,14 @@ python examples/offline_inference/pooling/embed_jina_embeddings_v3.py
 python examples/offline_inference/pooling/embed_matryoshka_fy.py
 ```
 
+## Named Entity Recognition (NER) usage
+
+```bash
+python examples/offline_inference/pooling/ner.py
+```
+
 ## Qwen3 reranker usage
 
 ```bash
-python qwen3_reranker.py
+python examples/offline_inference/pooling/qwen3_reranker.py
 ```

examples/offline_inference/pooling/ner.py

@@ -0,0 +1,54 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# Adapted from https://huggingface.co/boltuix/NeuroBERT-NER
+
+from argparse import Namespace
+
+from vllm import LLM, EngineArgs
+from vllm.utils import FlexibleArgumentParser
+
+
+def parse_args():
+    parser = FlexibleArgumentParser()
+    parser = EngineArgs.add_cli_args(parser)
+    # Set example specific arguments
+    parser.set_defaults(
+        model="boltuix/NeuroBERT-NER",
+        runner="pooling",
+        enforce_eager=True,
+        trust_remote_code=True,
+    )
+    return parser.parse_args()
+
+
+def main(args: Namespace):
+    # Sample prompts.
+    prompts = [
+        "Barack Obama visited Microsoft headquarters in Seattle on January 2025."
+    ]
+
+    # Create an LLM.
+    llm = LLM(**vars(args))
+    tokenizer = llm.get_tokenizer()
+    label_map = llm.llm_engine.vllm_config.model_config.hf_config.id2label
+
+    # Run inference
+    outputs = llm.encode(prompts)
+
+    for prompt, output in zip(prompts, outputs):
+        logits = output.outputs.data
+        predictions = logits.argmax(dim=-1)
+
+        # Map predictions to labels
+        tokens = tokenizer.convert_ids_to_tokens(output.prompt_token_ids)
+        labels = [label_map[p.item()] for p in predictions]
+
+        # Print results
+        for token, label in zip(tokens, labels):
+            if token not in tokenizer.all_special_tokens:
+                print(f"{token:15} → {label}")
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

examples/online_serving/pooling/README.md

@@ -12,6 +12,12 @@ python examples/online_serving/pooling/cohere_rerank_client.py
 python examples/online_serving/pooling/jinaai_rerank_client.py
 ```
 
+## Named Entity Recognition (NER) usage
+
+```bash
+python examples/online_serving/pooling/ner.py
+```
+
 ## Openai chat embedding for multimodal usage
 
 ```bash

examples/online_serving/pooling/ner.py

@@ -0,0 +1,71 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# Adapted from https://huggingface.co/boltuix/NeuroBERT-NER
+
+"""
+Example online usage of Pooling API for Named Entity Recognition (NER).
+
+Run `vllm serve <model> --runner pooling`
+to start up the server in vLLM. e.g.
+
+vllm serve boltuix/NeuroBERT-NER
+"""
+
+import argparse
+
+import requests
+import torch
+
+
+def post_http_request(prompt: dict, api_url: str) -> requests.Response:
+    headers = {"User-Agent": "Test Client"}
+    response = requests.post(api_url, headers=headers, json=prompt)
+    return response
+
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--host", type=str, default="localhost")
+    parser.add_argument("--port", type=int, default=8000)
+    parser.add_argument("--model", type=str, default="boltuix/NeuroBERT-NER")
+
+    return parser.parse_args()
+
+
+def main(args):
+    from transformers import AutoConfig, AutoTokenizer
+
+    api_url = f"http://{args.host}:{args.port}/pooling"
+    model_name = args.model
+
+    # Load tokenizer and config
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    config = AutoConfig.from_pretrained(model_name)
+    label_map = config.id2label
+
+    # Input text
+    text = "Barack Obama visited Microsoft headquarters in Seattle on January 2025."
+    prompt = {"model": model_name, "input": text}
+
+    pooling_response = post_http_request(prompt=prompt, api_url=api_url)
+
+    # Run inference
+    output = pooling_response.json()["data"][0]
+    logits = torch.tensor(output["data"])
+    predictions = logits.argmax(dim=-1)
+    inputs = tokenizer(text, return_tensors="pt")
+
+    # Map predictions to labels
+    tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
+    labels = [label_map[p.item()] for p in predictions]
+    assert len(tokens) == len(predictions)
+
+    # Print results
+    for token, label in zip(tokens, labels):
+        if token not in tokenizer.all_special_tokens:
+            print(f"{token:15} → {label}")
+
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)