[Feature]: support guided decoding parameters in the /chat/completions endpoint of trtllm-serve

🚀 The feature, motivation and pitch

openai.OpenAI.chat.completions.create doesn't have guided decoding fields in the method signature. Guided decoding fields like guided_json must be passed in extra_body.

If I make a request with extra_body, it fails with

openai.BadRequestError: Error code: 400 - {'object': 'error', 'message': "[{'type': 'extra_forbidden', 'loc': ('body', 'guided_json'), 'msg': 'Extra inputs are not permitted', 'input': {'type': 'object', 'properties': {'title': {'type': 'string'}, 'rating': {'type': 'number'}}, 'required': ['title', 'rating']}}]", 'type': 'BadRequestError', 'param': None, 'code': 400}

vllm serve and python -m sglang.launch_server APIs support extra_body and guided decoding parameters.

Steps to reproduce

1. Start container

set -u
export CONTAINER_NAME=extra-body-test
export IMG_NAME=nvcr.io/nvidia/tensorrt-llm/release:1.2.0rc4
docker run -it --rm --name=$CONTAINER_NAME \
  --gpus '"device=4"' \
  --shm-size=16GB \
  -e HF_TOKEN=$HF_TOKEN \
  -v "$HOME/trtllm-perf-isolation:/perf" \
  -u root \
  $IMG_NAME \
  /bin/bash

2. Deploy the model

trtllm-serve serve Qwen/Qwen3-0.6B --reasoning_parser deepseek-r1

3. Create a new shell, exec into the container and run the following script:

from openai import OpenAI
import sys

if len(sys.argv) > 1:
    model_name = sys.argv[1]
else:
    model_name = "gpt-oss-120b"
client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-used")
json_schema = {
    "type": "object",
    "properties": {
        "title": {
            "type": "string"
        },
        "rating": {
            "type": "number"
        }
    },
    "required": [
        "title",
        "rating"
    ]
}
confusing_json_schema = {
    "type": "object",
    "properties": {
        "name": {
            "type": "string"
        },
        "score": {
            "type": "number"
        }
    },
    "required": [
        "name",
        "score"
    ]
}
prompt = (f"Return the title and the rating based on the following movie review according to this JSON schema: {str(confusing_json_schema)}.\n"
          f"Review: Inception is a really well made film. I rate it four stars out of five.")
messages = [
    {"role": "user", "content": prompt},
]
response = client.chat.completions.create(
    model=model_name,
    messages=messages,
    extra_body={"guided_json": json_schema},
    stream=False
)
assistant_message = response.choices[0].message.content
print("message:")
print(assistant_message)
print("reasoning_content:")
print(getattr(response.choices[0].message, "reasoning_content", None))

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.