源码 examples/online_serving/openai_chat_completion_structured_outputs_with_reasoning.py
# SPDX-License-Identifier: Apache-2.0
"""
本示例展示如何使用推理模型（如 DeepSeekR1）生成结构化输出。
思考过程不会受到用户提供的 JSON 模式的引导，只有最终输出会被结构化。
要运行此示例，需要使用推理解析器启动 vLLM 服务器：
```bash
vllm serve deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B \
--enable-reasoning --reasoning-parser deepseek_r1
本示例演示如何使用 OpenAI Python 客户端库，从推理模型生成聊天补全内容。 """
from enum import Enum
from openai import OpenAI from pydantic import BaseModel
修改 OpenAI 的 API key 和 API base, 使用 vLLM 的 API 服务器。
openai_api_key = "EMPTY" openai_api_base = "http://localhost:8000/v1"
client = OpenAI( api_key=openai_api_key, base_url=openai_api_base, )
models = client.models.list() model = models.data[0].id
Guided decoding by Regex
使用 Regex 的引导式解码
prompt = ("What is the capital of France?")
completion = client.chat.completions.create( model=model, messages=[{ "role": "user", "content": prompt, }], extra_body={ "guided_regex": "(Paris|London)", }, ) print("reasoning_content: ", completion.choices[0].message.reasoning_content) print("content: ", completion.choices[0].message.content)
class People(BaseModel): name: str age: int
json_schema = People.model_json_schema()
prompt = ("Generate a JSON with the name and age of one random person.") completion = client.chat.completions.create( model=model, messages=[{ "role": "user", "content": prompt, }], extra_body={"guided_json": json_schema}, ) print("reasoning_content: ", completion.choices[0].message.reasoning_content) print("content: ", completion.choices[0].message.content)
使用 pydantic 模式的 JSON 引导式解码
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=model, messages=[{ "role": "user", "content": prompt, }], extra_body={"guided_json": json_schema}, ) print("reasoning_content: ", completion.choices[0].message.reasoning_content) print("content: ", completion.choices[0].message.content)
使用 Grammar 的引导式解码
simplified_sql_grammar = """ ?start: select_statement
?select_statement: "SELECT " column_list " FROM " table_name
?column_list: column_name ("," column_name)*
?table_name: identifier
?column_name: identifier
?identifier: /[a-zA-Z_][a-zA-Z0-9_]*/
"""
This may be very slow https://github.com/vllm-project/vllm/issues/12122
这可能很慢
prompt = ("Generate an SQL query to show the 'username' and 'email'" "from the 'users' table.") completion = client.chat.completions.create( model=model, messages=[{ "role": "user", "content": prompt, }], extra_body={"guided_grammar": simplified_sql_grammar}, ) print("reasoning_content: ", completion.choices[0].message.reasoning_content) print("content: ", completion.choices[0].message.content)