调用AWS Bedrock Claude Sonnet 3.5时图片JSON payload报错求助

调用AWS Bedrock Claude 3.5 Sonnet模型时，传入系统提示词、本地JPEG图片、文本上下文及查询语句后出现JSON payload错误，错误信息如下：

A client error occured: messages.0.content.1.image.source: Field required

相关代码如下：

def generate_conversation_bedrock(bedrock_client, model_id, input_text, input_image, max_tokens):

    response = client.invoke_model(
        modelId=model_id,
        body=json.dumps(
            {
                "anthropic_version": "bedrock-2023-05-31",
                "max_tokens": max_tokens,
                "system": sys_msg,
                "messages": [
                    {
                        "role": "user",
                        "content": [
                            {
                                "type": "text",
                                "text": input_text
                            },
                            {
                                "type": "image",
                                "image": {
                                    "format": "jpeg",
                                    "source": {
                                        "media_type": "image/jpeg",
                                        "data": input_image_base64  # Contiene la imagen en formato base64
                                    }
                                }
                            }
                        ]
                    }
                ],
            }
        ),
    )
    response_body = json.loads(response.get("body").read())
    return response_body

def main():
    model_id = "anthropic.claude-3-5-sonnet-20240620-v1:0"
    input_text = "describe la  imagen"
    input_image = "./combined_image.jpg"
    image = resize_and_compress_image(input_image)
    input_image_base64 = base64.b64encode(image).decode('utf-8')
   
    try:
        bedrock_client = boto3.client(service_name="bedrock-runtime", region_name="us-east-1")
        response = generate_conversation_bedrock(bedrock_client, model_id, input_text, input_image_base64, 1024)
        
        for output in response.get("content", []):
            print(output["text"])
     

    except ClientError as err:
        message = err.response['Error']['Message']
        print(f"A client error occured: {message}")

   

if __name__ == "__main__":
    main()

注：resize_and_compress_image函数用于将图片调整至模型输入限制尺寸

• JSON payload结构错误：Claude 3系列模型的图片内容格式不符合要求，错误地将source嵌套在image对象中。正确格式是直接在type: "image"的内容项中包含source字段，且source内必须指定type: "base64"。
• 函数参数不匹配：generate_conversation_bedrock函数定义的input_image参数未被使用，实际需要传入base64编码后的图片数据，需修改参数名为input_image_base64。
• 客户端对象误用：函数内使用了未定义的client变量，应替换为传入的bedrock_client参数。
• 系统提示词未定义：代码中引用的sys_msg变量未初始化，需补充定义系统提示词内容。

import json
import base64
import boto3
from botocore.exceptions import ClientError

# 假设resize_and_compress_image函数已正确实现
def resize_and_compress_image(image_path):
    # 此处为示例实现，需替换为你的实际逻辑
    with open(image_path, "rb") as f:
        return f.read()

def generate_conversation_bedrock(bedrock_client, model_id, input_text, input_image_base64, max_tokens, sys_msg):
    response = bedrock_client.invoke_model(
        modelId=model_id,
        body=json.dumps(
            {
                "anthropic_version": "bedrock-2023-05-31",
                "max_tokens": max_tokens,
                "system": sys_msg,
                "messages": [
                    {
                        "role": "user",
                        "content": [
                            {
                                "type": "text",
                                "text": input_text
                            },
                            {
                                "type": "image",
                                "source": {
                                    "type": "base64",
                                    "media_type": "image/jpeg",
                                    "data": input_image_base64
                                }
                            }
                        ]
                    }
                ],
            }
        ),
    )
    response_body = json.loads(response.get("body").read())
    return response_body

def main():
    model_id = "anthropic.claude-3-5-sonnet-20240620-v1:0"
    input_text = "describe la imagen"
    input_image = "./combined_image.jpg"
    # 定义系统提示词
    sys_msg = "你是一个专业的图片描述助手，需要根据用户提供的图片和文本指令输出准确的描述。"
    
    image = resize_and_compress_image(input_image)
    input_image_base64 = base64.b64encode(image).decode('utf-8')
   
    try:
        bedrock_client = boto3.client(service_name="bedrock-runtime", region_name="us-east-1")
        response = generate_conversation_bedrock(bedrock_client, model_id, input_text, input_image_base64, 1024, sys_msg)
        
        for output in response.get("content", []):
            print(output["text"])
     

    except ClientError as err:
        message = err.response['Error']['Message']
        print(f"A client error occured: {message}")

if __name__ == "__main__":
    main()

内容的提问来源于stack exchange，提问作者Diego