import gradio as gr
from transformers import AutoTokenizer, pipeline, AutoModelForSequenceClassification

# load the tokenizer and model from Hugging Face
tokenizer = AutoTokenizer.from_pretrained("ethanrom/a2")
model = AutoModelForSequenceClassification.from_pretrained("ethanrom/a2")

# define the classification labels
class_labels = ["Negative", "Positive", "No Impact", "Mixed"]

# create the zero-shot classification pipeline
classifier = pipeline("zero-shot-classification", model=model, tokenizer=tokenizer, device=0)

# define the Gradio interface
def predict_sentiment(text, model_choice):
    if model_choice == "bert":
        # use the default BERT sentiment analysis pipeline
        sentiment_classifier = pipeline("sentiment-analysis", device=0)
        result = sentiment_classifier(text)[0]
        label = result["label"]
        score = result["score"]
        return f"{label} ({score:.2f})"
    else:
        # use the fine-tuned RoBERTa model for multi-class classification
        labels = class_labels
        hypothesis_template = "This text is about {}."
        result = classifier(text, hypothesis_template=hypothesis_template, multi_class=True, labels=labels)
        scores = result["scores"]
        predicted_label = result["labels"][0]
        return f"{predicted_label} ({scores[0]:.2f})"

# define the Gradio interface inputs and outputs
inputs = [gr.inputs.Textbox(label="Input Text"), gr.inputs.Radio(["bert", "fine-tuned RoBERTa"], label="Model Choice")]
outputs = gr.outputs.Textbox(label="Sentiment Prediction")

# create the Gradio interface
gr.Interface(predict_sentiment, inputs, outputs, title="Sentiment Analysis App").launch()