import gradio as gr
from transformers import pipeline
import torch

# Initialize the zero-shot classification pipeline
try:
    classifier = pipeline(
        "zero-shot-classification",
        model="models/tasksource/ModernBERT-nli",
        device=0 if torch.cuda.is_available() else -1
    )
except Exception as e:
    print(f"Error loading model: {e}")
    classifier = None

def classify_text(text, candidate_labels):
    """
    Perform zero-shot classification on input text.
    
    Args:
        text (str): Input text to classify
        candidate_labels (str): Comma-separated string of possible labels
    
    Returns:
        list: List of (label, score) tuples
    """
    if classifier is None:
        # Return a default response when model fails to load
        return [("error", 1.0)]
    
    try:
        # Convert comma-separated string to list
        labels = [label.strip() for label in candidate_labels.split(",")]
        
        # Perform classification
        result = classifier(text, labels)
        
        # Convert results to list of (label, score) tuples
        return list(zip(result["labels"], result["scores"]))
    
    except Exception as e:
        print(f"Classification error: {e}")
        return [("error", 1.0)]

# Create Gradio interface
iface = gr.Interface(
    fn=classify_text,
    inputs=[
        gr.Textbox(
            label="Text to classify",
            placeholder="Enter text here...",
            value="all cats are blue"
        ),
        gr.Textbox(
            label="Possible labels (comma-separated)",
            placeholder="Enter labels...",
            value="true,false"
        )
    ],
    outputs=gr.Label(label="Classification Results"),
    title="Zero-Shot Text Classification",
    description="Classify text into given categories without any training examples.",
    examples=[
        ["all cats are blue", "true,false"],
        ["the sky is above us", "true,false"],
        ["birds can fly", "true,false,unknown"]
    ]
)

# Launch the app
if __name__ == "__main__":
    iface.launch(share=True)