[Bugfix][v1] xgrammar structured output supports Enum.

[chaunceyjiang]

FIX #15236

Structured output supports Enum.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

[chaunceyjiang]

Test whether xgrammar supports enum.

import xgrammar as xgr
from xgrammar.testing import  _is_grammar_accept_string

import json

# Guided decoding by JSON using Pydantic schema
class CarType(str, Enum):    # test Enum
    sedan = "sedan"
    suv = "SUV"
    truck = "Truck"
    coupe = "Coupe"


class CarDescription(BaseModel):
    brand: str
    model: str
    car_type: CarType



json_schema = CarDescription.model_json_schema()

instance = CarDescription(brand="Toyota", model="Corolla", car_type="sedan")

json_schema_grammar = xgr.Grammar.from_json_schema(json.dumps(json_schema))

# instance: pydantic model, json string, or any other object (dumped to json string)
if isinstance(instance, BaseModel):
    instance = json.dumps(instance.model_dump(mode="json", round_trip=True))



assert _is_grammar_accept_string(json_schema_grammar, instance)

[chaunceyjiang]

test

vllm serve Qwen/Qwen2.5-3B-Instruct

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="-",
)


class CarType(str, Enum):
    sedan = "sedan"
    suv = "SUV"
    truck = "Truck"
    coupe = "Coupe"


class CarDescription(BaseModel):
    brand: str
    model: str
    car_type: CarType
json_schema = CarDescription.model_json_schema()
prompt = ("Generate a JSON with the brand, model and car_type of"
          "the most iconic car from the 90's")
completion = client.chat.completions.create(
    model="Qwen/Qwen2.5-3B-Instruct",
    messages=[{
        "role": "user",
        "content": prompt,
    }],
    extra_body={"guided_json": json_schema},
)
print(completion.choices[0].message.content)

# python test.py 
[2025-03-27 03:24:41] INFO _client.py:1026: HTTP Request: POST http://localhost:8000/v1/chat/completions "HTTP/1.1 200 OK"
{
  "brand": "Ford",
  "model": "Mustang",
  "car_type": "SUV"
}

[chaunceyjiang]

/cc @aarnphm @agm-eratosth @subnet-dev @DarkLight1337 PTAL.

[ItzAmirreza]

Much love

[aarnphm]

Can you add a test for this. Thanks.

[chaunceyjiang]

@DarkLight1337 Hi, Can you help add a ready label for e2e?

[aarnphm]

Please fix the tests accordingly first.

[DarkLight1337]

Approving to unblock merge once the tests pass

[DarkLight1337]

Test passes

[russellb]

It looks like full CI didn’t run on this PR for some reason before it was merged. It broke a structured output test for V1. See the failure in #15718

[DarkLight1337]

Oh, my bad. I saw the entrypoints tests pass but forgot about V1.

[chaunceyjiang]

@russellb Thanks～

[DarkLight1337]

Are you able to fix this quickly?

[chaunceyjiang]

Okay, I'll check it out right away.