下面示例是使用langchain调用多模态的示例

数据处理

import os
import base64
from mimetypes import guess_type

def local_image_to_data_url(image_path):
    # Guess the MIME type of the image based on the file extension
    mime_type, _ = guess_type(image_path)
    if mime_type is None:
        mime_type = 'application/octet-stream'  # Default MIME type if none is found

    # Read and encode the image file
    with open(image_path, "rb") as image_file:
        base64_encoded_data = base64.b64encode(image_file.read()).decode('utf-8')

    # Construct the data URL
    return f"data:{mime_type};base64,{base64_encoded_data}"


def get_image_message(img_path):
    def _get_template(image_path):
        print('文件路径为：' + image_path)
        url = local_image_to_data_url(image_path)
        return {
            "type": "image_url",
            "image_url": {
                "url": url
            },
        }

    if not os.path.exists(img_path):
        print(f"Error: Path {img_path} does not exist.")
        return

    image_extensions = ['.png', '.jpg', '.jpeg']
    image_message = []
    # 判断传入的是一个file还是folder
    if os.path.isfile(img_path):
        image_url = _get_template(img_path)
        image_message.append(image_url)
    elif os.path.isdir(img_path):
        for f in os.listdir(img_path):
            if any(f.lower().endswith(ext) for ext in image_extensions):
                image_url = _get_template(os.path.join(img_path, f))
                image_message.append(image_url)
    else:
        print("错误！请检测传入的图片文件夹目录结构。")

    return image_message


if __name__ == '__main__':
    img_path = "img/temp/"
    user_image_message = get_image_message(img_path)

主程序

import os
from dotenv import load_dotenv, find_dotenv

_ = load_dotenv(find_dotenv())

from langchain.prompts import ChatPromptTemplate
from langchain_core.output_parsers import PydanticOutputParser
from langchain.schema.runnable import RunnablePassthrough
from pydantic import BaseModel, Field, validator
from langchain_openai import AzureChatOpenAI

from ulits.data_processing import *


class Result(BaseModel):
    Observation: str = Field(description="仔细观察你收的图像, 详细描述图像中的内容，用中中文回答")
    Thought: str = Field(description="根据你的描述，结合用户的视频搜索提示词，思考当前画面是否是用户想要寻找的画面")
    Result: bool = Field(description="给出结论，根据你的思考判断当前画面是否是用户想要搜索的画面", default=False)


# OutputParser
parser = PydanticOutputParser(pydantic_object=Result)

system_prompt_path = "prompt/prompt1.txt"
with open(system_prompt_path, 'r', encoding='utf-8') as f:
    system_prompt = f.read()

user_text_message = [
    {
        "type": "text",
        "text": "用户输入的视频搜索提示词：\n{query}\n"
    }]

img_path = "img/"
user_image_message = get_image_message(img_path)

user_message = user_text_message + user_image_message

# Prompt 模板
prompt = ChatPromptTemplate.from_messages(
    [
        ("system", system_prompt),
        ("user", user_message),
    ]
).partial(format_instructions=parser.get_format_instructions())

# 模型
model = AzureChatOpenAI(azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
                        azure_deployment=os.environ["AZURE_OPENAI_DEPLOYMENT_NAME"],
                        openai_api_version=os.environ["AZURE_OPENAI_API_VERSION"],
                        temperature=0,
                        model_kwargs={"seed": 42})

# LCEL 表达式
runnable = (
        {"query": RunnablePassthrough()} | prompt | model | parser
)

# 运行
print(runnable.invoke("搜索一个拿伞的女人"))

print("end")