app.py

import gradio as gr
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline
from deep_translator import GoogleTranslator
import torch
import string

def preprocess_data(text: str) -> str:
    return text.lower().translate(str.maketrans("", "", string.punctuation)).strip()

SARCASTIC_MODEL_PATH = "helinivan/english-sarcasm-detector"
SENTIMENT_MODEL_PATH = "lxyuan/distilbert-base-multilingual-cased-sentiments-student"

sarcasm_tokenizer = AutoTokenizer.from_pretrained(SARCASTIC_MODEL_PATH)
sarcasm_model = AutoModelForSequenceClassification.from_pretrained(SARCASTIC_MODEL_PATH)
sentiment_analyzer = pipeline("text-classification", model=SENTIMENT_MODEL_PATH, return_all_scores=True)

def analyze_text(user_input):
    
    translated_text = GoogleTranslator(source="auto", target="en").translate(user_input)
    preprocessed_text = preprocess_data(translated_text)
    tokenized_text = sarcasm_tokenizer([preprocessed_text], padding=True, truncation=True, max_length=256, return_tensors="pt")
    with torch.no_grad():
        output = sarcasm_model(**tokenized_text)
    probs = torch.nn.functional.softmax(output.logits, dim=-1).tolist()[0]
    sarcasm_confidence = max(probs)
    is_sarcastic = probs.index(sarcasm_confidence)
    if is_sarcastic:
        return f"Sarcastic"
    else:
        sentiment_scores = sentiment_analyzer(translated_text)[0]
        sentiment_result = max(sentiment_scores, key=lambda x: x["score"])
        return f"{sentiment_result['label'].capitalize()}"
    
iface = gr.Interface(
    fn=analyze_text,
    inputs=gr.Textbox(label="Enter your text (Tamil)"),
    outputs=gr.Textbox(label="Analysis Result"),
    description="Enter text in TAMIL",
    examples=[
        ['ஹாய் மாலினி, நான் இதை சொல்லியே ஆகணும், நீ அவ்ளோ அழகு, இங்க உன்னைவிட ஒரு அழகா யாரும் பாத்துருக்க மாட்டாங்க'],
        ['இது நல்ல இல்ல'],
        ['நம்ம ஜெயிச்சிட்டோம் மாறா! ']
    ],
)

iface.launch(share=True)