app.py

import torch
from transformers import MBartForConditionalGeneration, AutoTokenizer, AutoModelForCausalLM, pipeline
import gradio as gr
import requests
import io
from PIL import Image
import os

# Set up the Hugging Face API key from environment variables
hf_api_key = os.getenv("new_hf_token")
if not hf_api_key:
    raise ValueError("Hugging Face API key not found! Please set the 'HF_API_KEY' environment variable.")
headers = {"Authorization": f"Bearer {hf_api_key}"}

# Define the text-to-image model URL
API_URL = "https://api-inference.huggingface.co/models/CompVis/stable-diffusion-v1-4"

# Use AutoTokenizer to avoid tokenizer mismatch warnings
translation_model_name = "facebook/mbart-large-50-many-to-one-mmt"
tokenizer = AutoTokenizer.from_pretrained(translation_model_name)  # Use AutoTokenizer to avoid warnings
translation_model = MBartForConditionalGeneration.from_pretrained(translation_model_name)

# Load a text generation model from Hugging Face using accelerate for memory optimization
text_generation_model_name = "EleutherAI/gpt-neo-2.7B"
text_tokenizer = AutoTokenizer.from_pretrained(text_generation_model_name)
text_model = AutoModelForCausalLM.from_pretrained(
    text_generation_model_name, 
    device_map="auto", 
    torch_dtype=torch.float32
)

# Create a pipeline for text generation
text_generator = pipeline("text-generation", model=text_model, tokenizer=text_tokenizer)

# Function to generate an image using Hugging Face's text-to-image model
def generate_image_from_text(translated_text):
    try:
        response = requests.post(API_URL, headers=headers, json={"inputs": translated_text})
        if response.status_code != 200:
            return None, f"Error generating image: {response.text}"
        
        image_bytes = response.content
        image = Image.open(io.BytesIO(image_bytes))
        return image, None
    except Exception as e:
        return None, f"Error during image generation: {e}"

# Define the function to translate Tamil text, generate an image, and create a descriptive text
def translate_generate_image_and_text(tamil_text):
    try:
        tokenizer.src_lang = "ta_IN"
        inputs = tokenizer(tamil_text, return_tensors="pt")
        translated_tokens = translation_model.generate(**inputs, forced_bos_token_id=tokenizer.lang_code_to_id["en_XX"])
        translated_text = tokenizer.batch_decode(translated_tokens, skip_special_tokens=True)[0]
    except Exception as e:
        return f"Error during translation: {e}", None, None

    try:
        image, error_message = generate_image_from_text(translated_text)
        if error_message:
            return translated_text, None, error_message
    except Exception as e:
        return translated_text, None, f"Error during image generation: {e}"

    try:
        descriptive_text = text_generator(translated_text, max_length=100, num_return_sequences=1, temperature=0.7, top_p=0.9)[0]['generated_text']
    except Exception as e:
        return translated_text, image, f"Error during text generation: {e}"

    return translated_text, image, descriptive_text

# Gradio interface setup
iface = gr.Interface(
    fn=translate_generate_image_and_text,
    inputs=gr.Textbox(lines=2, placeholder="Enter Tamil text here..."),
    outputs=[gr.Textbox(label="Translated English Text"), 
             gr.Image(label="Generated Image"),
             gr.Textbox(label="Generated Descriptive Text")],
    title="Tamil to English Translation, Image Creation, and Descriptive Text Generation",
    description="Translate Tamil text to English using Facebook's mbart-large-50 model, create an image using the translated text, and generate a descriptive text based on the translated content.",
)

# Launch the Gradio app
iface.launch()