Vladmun1337/multiclip-classifier-ro

MultiClip human written vs machine generated review classifier (Romanian)

This is a late fusion variant of the multilingual CLIP classifier, inspired by sentence transformers.

Usage

Use code below, with a text example and an image. GPU acceleration is recommended.

import torch
from PIL import Image
from transformers import AutoTokenizer, AutoProcessor, AutoModelForSequenceClassification

# 1. Download
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = AutoModelForSequenceClassification.from_pretrained("Vladmun1337/multiclip-classifier-ro", trust_remote_code=True).to(device)
model.eval()

# 2. Tokenizers
tokenizer = AutoTokenizer.from_pretrained("sentence-transformers/clip-ViT-B-32-multilingual-v1")
processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")

# 3. Sample input
text = ["Am cumpărat acest frigider ieri și sunt foarte mulțumit. Recomand!"]
image = Image.open("example.png").convert("RGB")

text_inputs = tokenizer(text, padding='max_length', truncation=True, max_length=256, return_tensors='pt')
image_inputs = processor(images=image, return_tensors="pt")

# 4. Inference
with torch.no_grad():
    logits = model(
        input_ids=text_inputs['input_ids'].to(device), 
        attention_mask=text_inputs['attention_mask'].to(device), 
        pixel_values=image_inputs['pixel_values'].to(device)
    )
    predicted_class = torch.argmax(logits, dim=1).item()


labels = {0: "Human", 1: "AI"}
print(f"Prediction: {labels[predicted_class]}")