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

# Load the fine-tuned model and tokenizer
model_name = "ethanrom/a2"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Load the pretrained model and tokenizer
pretrained_model_name = "roberta-large-mnli"
pretrained_tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name)
pretrained_model = pipeline("zero-shot-classification", model=pretrained_model_name, tokenizer=pretrained_tokenizer)
candidate_labels = ["negative", "positive", "no impact", "mixed"]


def predict_sentiment(text_input, model_selection):
    if model_selection == "Fine-tuned":
        # Use the fine-tuned model
        inputs = tokenizer.encode_plus(text_input, return_tensors='pt')
        outputs = model(**inputs)
        logits = outputs.logits.detach().cpu().numpy()[0]
        predicted_class = int(logits.argmax())
        return candidate_labels[predicted_class]
    else:
        # Use the pretrained model
        result = pretrained_model(text_input, candidate_labels)
        predicted_class = result["labels"][0]
        return predicted_class

inputs = [
    gr.inputs.Textbox("Enter text"),
    gr.inputs.Dropdown(["Pretrained", "Fine-tuned"], label="Select model"),
]

outputs = gr.outputs.Textbox(label="Predicted Sentiment")

gr.Interface(fn=predict_sentiment, inputs=inputs, outputs=outputs, title="Sentiment Analysis", description="Compare the output of two models", examples=[
        ["max laid his hand upon the old man's arm", "Pretrained Model"],
        ["the red sword sealed their vows!", "Fine-tuned Model"],
        ["and that is why, the lonesome day,", "Pretrained Model"],
        ["it flows so long as falls the rain", "Fine-tuned Model"],
        ["thy hands all cunning arts that women prize", "Pretrained Model"],
        ["on us lift up the light", "Fine-tuned Model"],
    ],).launch();