# import gradio as gr
# gr.load("models/karthikvarunn/skar_propaganda_v1.1").launch()

# import gradio as gr
# from transformers import pipeline
 
# # Load the model correctly by specifying the repo id in the correct format
# classifier = pipeline("zero-shot-classification", model="karthikvarunn/skar_propaganda_v1.1")

# def format_output(result):
#     # Unpack the result
#     sequence = result['sequence']
#     labels = result['labels']
#     scores = result['scores']
#     # Create a formatted string or list to display results clearly
#     output_lines = []
#     for label, score in zip(labels, scores):
#         # Strip any leading/trailing whitespace from the label
#         label = label.strip()
#         output_lines.append(f"{label}: {score:.2f}")  # Formatting score to two decimal places
#     return "\n".join(output_lines)
 
# def predict(text, classes):
#     labels = [label.strip() for label in classes.split(",")]  # Ensure labels are properly stripped of whitespace
#     result = classifier(text, labels)
#     formatted_result = format_output(result)
#     return formatted_result


# demo = gr.Interface(
#     fn=predict,
#     inputs=[
#         gr.Textbox(lines=2, placeholder="Type something here...", value="This miracle drug can save lives!"),
#         gr.Textbox(lines=2, placeholder="Possible labels, comma-separated", value="Doubt, Loaded Language, Jingoism, Hyperbole, no_propaganda")
#     ],
#     outputs=gr.Textbox(label="Formatted Output"),
#     title="Text Classification",
#     description="Enter text and labels to classify the text."
# )

 
# demo.launch()

import gradio as gr
from transformers import pipeline
 
# Initialize the zero-shot classification pipeline with your model from Hugging Face
classifier = pipeline("zero-shot-classification", model="karthikvarunn/skar_propaganda_v1.1")
 
def predict(text, classes):
    # Split the comma-separated labels provided by the user and strip whitespace
    labels = [label.strip() for label in classes.split(",")]
    # Use the classifier to get predictions
    result = classifier(text, labels)
    # Convert labels and scores to a dictionary for the Gradio Label component
    label_scores = {label: score for label, score in zip(result['labels'], result['scores'])}
    return label_scores
 
# Create the Gradio interface
demo = gr.Interface(
    fn=predict,
    inputs=[
        gr.Textbox(lines=2, placeholder="Type a sentence to find the propaganda technique...", value="This miracle drug can save lives!"),
        gr.Textbox(lines=2, placeholder="Possible labels, comma-separated", value="Doubt, Loaded Language, Jingoism, Hyperbole, no_propaganda")
    ],
    outputs=gr.Label(num_top_classes=None),  # Automatically render scores as a bar chart
    title="Zero-shot Classification for propaganda techniques",
    description="Enter text and labels to classify the text."
)
 
# Launch the Gradio app
demo.launch()