app.py

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import gradio as gr

model_name = "Helsinki-NLP/opus-mt-en-ur"

print("Downloading tokenizer and model directly...")
# Loading the tokenizer and model directly bypasses the buggy pipeline
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
print("Model loaded successfully!")

def translate_en_to_ur(text):
    if not text.strip():
        return ""
    
    # 1. Tokenize: Convert the text into numbers the model understands
    inputs = tokenizer(text, return_tensors="pt", padding=True)
    
    # 2. Generate: The model predicts the Urdu translation tokens
    translated_tokens = model.generate(**inputs)
    
    # 3. Decode: Convert the predicted tokens back into readable text
    result = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
    return result

# Build the Gradio Interface
interface = gr.Interface(
    fn=translate_en_to_ur,
    inputs=gr.Textbox(lines=5, placeholder="Enter English text here...", label="English Input"),
    outputs=gr.Textbox(lines=5, label="Urdu Translation"),
    title="English to Urdu Translator 🇵🇰",
    description="Enter any English text and the AI will translate it into Urdu.",
    theme="default"
)

# Launch the app
interface.launch(share=True)