This is part of a YouTube tutorial series by MakeOps.
The file is a script that invokes Qwen2.5 VL via Amazon Bedrock to describe and image.
usage: main.py [-h] image
positional arguments:
image
options:
-h, --help show this help message and exit
"boto3>=1.40.6",
"dotenv>=0.9.9",
"pillow>=11.3.0",
"torchvision>=0.23.0",
"transformers>=4.55.0",| import json | |
| import os | |
| import base64 | |
| import boto3 | |
| from io import BytesIO | |
| from argparse import ArgumentParser | |
| from dotenv import load_dotenv | |
| from PIL import Image | |
| from transformers import AutoProcessor | |
| load_dotenv() | |
| bedrock = boto3.client('bedrock-runtime', region_name='us-east-1') | |
| QWEN_MODEL_ID = os.environ.get('QWEN_MODEL_ID') | |
| def image_to_base64(img: Image, img_format='JPEG'): | |
| '''Convert PIL image to Base64''' | |
| buffer = BytesIO() | |
| img.save(buffer, format=img_format) | |
| img_bytes = buffer.getvalue() | |
| return base64.b64encode(img_bytes).decode() | |
| def invoke_qwen(prompt: str, image_path: str, temp=0.0, max_tokens=2048, top_p=0.9): | |
| processor = AutoProcessor.from_pretrained('Qwen/Qwen2.5-VL-7B-Instruct', use_fast=True) | |
| messages = [ | |
| { | |
| "role": "user", | |
| "content": [ | |
| { "type": "image", "image": f"file://{image_path}" }, | |
| { "type": "text", "text": prompt }, | |
| ], | |
| } | |
| ] | |
| prompt = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True) | |
| response = bedrock.invoke_model( | |
| modelId=QWEN_MODEL_ID, | |
| body=json.dumps({ | |
| 'prompt': prompt, | |
| 'temperature': temp, | |
| 'max_gen_len': max_tokens, | |
| 'top_p': top_p, | |
| 'images': [image_to_base64(Image.open(image_path))] | |
| }), | |
| accept='application/json', | |
| contentType='application/json' | |
| ) | |
| response_body = json.load(response['body']) | |
| return response_body['choices'] | |
| def main(): | |
| parser = ArgumentParser() | |
| parser.add_argument('image') | |
| args = parser.parse_args() | |
| print(f'Using Qwen to Process Image modelId={QWEN_MODEL_ID}') | |
| full_path = os.path.join(os.path.dirname(__file__), args.image) | |
| print('full path', full_path) | |
| response = invoke_qwen('Explain what you see in this image', full_path) | |
| for response_item in response: | |
| print('qwen >', response_item['text']) | |
| if __name__ == "__main__": | |
| exit(main()) |