shilinxu/Qwen2-VL-2B-ViT

import torch
import requests
from PIL import Image
from transformers import AutoModel, AutoProcessor, AutoImageProcessor
from transformers import Qwen2VLForConditionalGeneration
image_processor = AutoImageProcessor.from_pretrained("shilinxu/Qwen2-VL-2B-ViT", trust_remote_code=True)
vit = AutoModel.from_pretrained("shilinxu/Qwen2-VL-2B-ViT", trust_remote_code=True, device_map='auto',torch_dtype=torch.bfloat16, attn_implementation='flash_attention_2')

url = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg"
image = Image.open(requests.get(url, stream=True).raw).convert("RGB")
images = [image]

inputs = image_processor(images, return_tensors="pt")
pixel_values = inputs['pixel_values'].to(device=vit.device, dtype=vit.dtype)
image_grid_thw = inputs['image_grid_thw']
image_embeds = vit(pixel_values, grid_thw=image_grid_thw)