gradio_postgres_workflow.py

# -*- coding: utf-8 -*-

"""
Postgres Workflow

使用定制的 Workflow 查询 Postgres 数据库

使用方法:
1. 下载 Qwen3 模型
2. 启动 vLLM 服务：
    cd test_qwen3
    bash vllm_server.sh
3. 启动 Gradio 应用：
    python gradio_postgres_workflow.py
4. 打开浏览器访问：
    http://localhost:7860/
5. 可以尝试以下问题：
    - 告诉我用户编号 103 的用户的订单信息
    - 查一下该用户订单的物流状态
"""

from gradio_ui import create_ui
from postgres_agent import PGAgent
from postgres_workflow import PGWorkflow


# LLM 配置
LLM_CFG = {
    'model': 'Qwen3-0.6B-FP8',
    'model_server': 'http://localhost:8000/v1',
    'api_key': 'token-kcgyrk',
    'generate_cfg': {
        'top_p': 0.95,
        'temperature': 0.6,
    }
}


# Postgres 数据库配置
DB_CONFIG = {
    "host": "localhost",
    "port": "5432",
    "database": "ecommerce_orders",
    "user": "admin",
    "password": "admin-password",
}


def create_react_agent(llm_cfg, db_config):
    # 实例化 Workflow
    pga = PGAgent(llm_cfg, db_config)
    react_agent = pga.create_react_agent()
    return react_agent


pgwf = PGWorkflow(LLM_CFG, DB_CONFIG)
my_bot = create_react_agent(LLM_CFG, DB_CONFIG)


def generate_response(message, history, max_history=4):
    if not message.strip():
        return message, history

    messages = [{'role': 'user', 'content': message}]

    # 保留最后 max_history 条历史记录
    messages = history[-max_history:] + messages

    history.append({"role": "user", "content": message})
    history.append({"role": "assistant", "content": "工作流运行中 ..."})

    # 注入 context 后的 messages
    messages = pgwf.workflow(messages)

    # 流式响应
    for chunk in my_bot.run(messages):
        content = chunk[-1].get("content", "")
        history[-1]["content"] = content
        yield "", history

    yield "", history


if __name__ == "__main__":
    model_name = LLM_CFG['model']
    demo = create_ui(llm_func=generate_response,
                     tab_name="Gradio APP - Postgres Workflow",
                     main_title="Postgres Workflow Demo",
                     sub_title=f"{model_name}")
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        show_api=False,
        share=False  # 内部使用时，必须为 False
    )