app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AlbertForSequenceClassification
import numpy as np
import os
import gdown

# Define Google Drive folder IDs for each model
model_drive_ids = {
    "sentiment": "your_sentiment_folder_id",  # Replace with actual folder ID
    "emotion": "your_emotion_folder_id",      # Replace with actual folder ID
    "hate_speech": "your_hate_speech_folder_id",  # Replace with actual folder ID
    "sarcasm": "your_sarcasm_folder_id"       # Replace with actual folder ID
}

# Define local directory to store downloaded models
save_dir = "./saved_models"
os.makedirs(save_dir, exist_ok=True)

# Download models from Google Drive
for task, folder_id in model_drive_ids.items():
    output_dir = os.path.join(save_dir, task)
    if not os.path.exists(output_dir):
        gdown.download_folder(
            f"https://drive.google.com/drive/folders/1kEXKoJxxD5-0FO8WvtagzseSIC5q-rRY?usp=sharing/{folder_id}",
            output=output_dir,
            quiet=False
        )

# Define model paths
tasks = ["sentiment", "emotion", "hate_speech", "sarcasm"]
model_paths = {task: f"{save_dir}/{task}" for task in tasks}

# Define label mappings
label_mappings = {
    "sentiment": ["negative", "neutral", "positive"],
    "emotion": ["happy", "sad", "angry", "fear"],
    "hate_speech": ["no", "yes"],
    "sarcasm": ["no", "yes"]
}

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("ai4bharat/indic-bert")

# Load all models
models = {}
for task in tasks:
    if not os.path.exists(model_paths[task]):
        raise FileNotFoundError(f"Model directory {model_paths[task]} not found.")
    models[task] = AlbertForSequenceClassification.from_pretrained(model_paths[task])

# Function to predict for a single task
def predict_task(text, task, model, tokenizer, max_length=128):
    inputs = tokenizer(
        text,
        padding=True,
        truncation=True,
        max_length=max_length,
        return_tensors="pt"
    )
    
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probabilities = torch.softmax(logits, dim=1).squeeze().cpu().numpy()
    
    labels = label_mappings[task]
    return {label: f"{prob*100:.2f}%" for label, prob in zip(labels, probabilities)}

# Gradio interface function
def predict_all_tasks(text):
    if not text.strip():
        return "Please enter some text."
    
    results = {}
    for task in tasks:
        results[task] = predict_task(text, task, models[task], tokenizer)
    
    output = ""
    for task, probs in results.items():
        output += f"\n{task.capitalize()} Prediction:\n"
        for label, prob in probs.items():
            output += f"  {label}: {prob}\n"
    
    return output

# Create Gradio interface
iface = gr.Interface(
    fn=predict_all_tasks,
    inputs=gr.Textbox(lines=2, placeholder="Enter Telugu text here..."),
    outputs="text",
    title="Telugu Text Analysis",
    description="Enter Telugu text to predict sentiment, emotion, hate speech, and sarcasm."
)

if __name__ == "__main__":
    iface.launch(server_name="0.0.0.0", server_port=7860)