AI合同单据识别-自定义字段信息抽取-小帮手更新

# %%import osfrom openai import OpenAIimport base64import pandas as pdimport json# %%client = OpenAI(    # 若没有配置环境变量，请用百炼API Key将下行替换为：api_key="sk-xxx",    api_key="",    base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",)#  Base64 编码格式def encode_image(image_path):    with open(image_path, "rb") as image_file:        return base64.b64encode(image_file.read()).decode("utf-8")# %%def extract_info(img_path):    base64_image = encode_image(img_path)    completion = client.chat.completions.create(        model="qwen-omni-turbo-latest",        # model="qwen-vl-max-latest",        messages=[            {                "role": "system",                "content": """你需要提取出买方、卖方、合同号、签订日期、产品、规格型号、数量、含税单价、不含税总额、交货时间、交货方式、交货地点，请输出JSON 字符串，不要输出其它无关内容。                示例：                ```json                    {                        "买方": "???",                        "卖方": "???",                        "合同号": "???",                        "签订日期": "???",                        "产品": "???",                        "规格型号": "???",                        "数量": "???",                        "含税单价": "???",                        "不含税总额": "???",                        "交货时间": "???",                        "交货方式": "???",                        "交货地点": "???"                    }                    ```                """,            },            {                "role": "user",                "content": [                    {                        "type": "image_url",                        "image_url": {"url": f"data:image/png;base64,{base64_image}"},                    }                ],            },        ],        # 设置输出数据的模态，当前支持两种：["text","audio"]、["text"]        modalities=["text"],        response_format={"type": "json_object"},        # stream 必须设置为 True，否则会报错        stream=True,    )    content = ""    for chunk in completion:        # print(chunk.choices[0].delta)        if chunk.choices[0].delta.content:            content += chunk.choices[0].delta.content    # print(content)    return json.loads(content)# %%record_list = []for root, dirs, files in os.walk("./imgs"):    for file in files:        print(os.path.join(root, file))        res = extract_info(os.path.join(root, file))        res["文件名"] = file        record_list.append(res)# %%df = pd.DataFrame(record_list)df.to_csv("data.csv", index=False)