kosmos.py

from transformers import AutoProcessor, AutoModelForVision2Seq
from PIL import Image
from os import path
from torchvision.transforms import ToTensor

model_id = "microsoft/kosmos-2-patch14-224"
model = AutoModelForVision2Seq.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)

def single_image_classification(image, prompt="", max_new_tokens=30):
    inputs = processor(text=prompt, images=image, return_tensors="pt")
    
    generated_ids = model.generate(
        pixel_values=inputs["pixel_values"],
        input_ids=inputs["input_ids"],
        attention_mask=inputs["attention_mask"],
        image_embeds=None,
        image_embeds_position_mask=inputs["image_embeds_position_mask"],
        use_cache=True,
        max_new_tokens=max_new_tokens,
    )
    generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

    print("GENERATED:", generated_text)

    processed_text = processor.post_process_generation(generated_text, cleanup_and_extract=False)

    print("PROCESSED:", processed_text)
    return processed_text