import torch
from PIL import Image
from transformers import AutoProcessor, AutoModelForImageClassification
import gradio as gr

# Load model and processor
from transformers import AutoModelForImageClassification, AutoProcessor

model = AutoModelForImageClassification.from_pretrained("model")
processor = AutoProcessor.from_pretrained("model")

model = AutoModelForImageClassification.from_pretrained(model_name)
processor = AutoProcessor.from_pretrained(model_name)

labels = model.config.id2label  # e.g., {0: "non-hateful", 1: "hateful"}

def classify_meme(image: Image.Image):
    inputs = processor(images=image, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model(**inputs)
        probs = torch.nn.functional.softmax(outputs.logits, dim=-1)
    predictions = {labels[i]: float(probs[0][i]) for i in range(len(labels))}
    return predictions

# Gradio interface
demo = gr.Interface(
    fn=classify_meme,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=2),
    title="Meme Sentiment Classifier (SigLIP2)",
    description="Upload a meme to classify its sentiment using a SigLIP2-based model."
)
from transformers import AutoModelForImageClassification, AutoProcessor

model = AutoModelForImageClassification.from_pretrained("model")
processor = AutoProcessor.from_pretrained("model")

if __name__ == "__main__":
    demo.launch()