import gradio as gr
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
from peft import PeftModel, PeftConfig

base_model = "cardiffnlp/twitter-roberta-base-sentiment-latest"
adapter_model = 'saideep-arikontham/twitter-roberta-base-sentiment-latest-trump-stance-1'

# define label maps
id2label = {0: "Anti-Trump", 1 : "Pro-Trump"}
label2id = {"Anti-Trump" : 0, "Pro-Trump" : 1}

# generate classification model from model_checkpoint
model = AutoModelForSequenceClassification.from_pretrained(base_model, num_labels=2, id2label = id2label, label2id = label2id, ignore_mismatched_sizes=True)

model = PeftModel.from_pretrained(model, adapter_model)
tokenizer = AutoTokenizer.from_pretrained(adapter_model)

def greet(text):
    
    model.to('cpu')
    inputs = tokenizer.encode(text, return_tensors="pt").to("cpu")
    # compute logits
    logits = model(inputs).logits
    # convert logits to label
    predictions = torch.argmax(logits)

    return "This text is " + id2label[predictions.tolist()] + "!!"



demo = gr.Interface(fn=greet, inputs="text", outputs="text")
demo.launch()