amoe-ultrasparse

import torch
from PIL import Image
from transformers import AutoModel, AutoImageProcessor

model_id = "tiiuae/amoe-ultrasparse"
model = AutoModel.from_pretrained(model_id, trust_remote_code=True).to("cuda", dtype=torch.bfloat16)
processor = AutoImageProcessor.from_pretrained(model_id, trust_remote_code=True)

image = Image.open("image.jpg").convert("RGB")
inputs = processor(image, return_tensors="pt").to("cuda")
inputs["pixel_values"] = inputs["pixel_values"].to(torch.bfloat16)

with torch.no_grad():
    outputs = model(**inputs)

# Options: 'amoe' (768d), 'siglip2' (1152d), 'dinov3' (1024d)
patch_features = outputs["patch_features"]["amoe"]         # (Batch, Tokens, 768)
summary_features = outputs["summary_features"]["siglip2"]  # (Batch, 1152)