# app.py - Final Version with English Comments and UI Text
import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# --- 1. Define model and tokenizer paths ---
model_name = "doganbilir/mt5-Turkish-English-Summarizer" 
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# --- 2. Load model and tokenizer from the Hub ---
try:
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
    print("Model and tokenizer loaded successfully!")
except Exception as e:
    print(f"An error occurred while loading the model: {e}")
    model = None

# --- 3. Define the summarization function ---
def summarize(text, language_choice, min_length):
    if model is None:
        return "Model could not be loaded. Please check the Space logs."
    if not text or not text.strip():
        return "Please enter text to summarize."
    
    # Set the prompt based on the language choice
    if language_choice == "Turkish Text -> Turkish Summary":
        prompt = f"summarize: {text}"
    elif language_choice == "English Text -> English Summary":
        prompt = f"summarize: {text}"
    else: # TR -> EN (Experimental)
        prompt = f"summarize Turkish to English: {text}"
        
    try:
        inputs = tokenizer(prompt, return_tensors="pt", max_length=1024, truncation=True).to(device)
        
        summary_ids = model.generate(
            input_ids=inputs.input_ids,
            attention_mask=inputs.attention_mask,
            max_new_tokens=150,
            num_beams=5,
            min_new_tokens=int(min_length), # Use the value from the slider
            early_stopping=True,
            no_repeat_ngram_size=3
        )
        
        summary = tokenizer.batch_decode(summary_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)[0]
        return summary.strip()
    except Exception as e:
        return f"An error occurred during summarization: {str(e)}"

# --- 4. Create the Gradio Interface ---
iface = gr.Interface(
    fn=summarize,
    inputs=[
        gr.Textbox(lines=15, placeholder="Paste the text you want to summarize here...", label="Text Input"),
        gr.Radio(
            ["Turkish Text -> Turkish Summary", "English Text -> English Summary", "Turkish Text -> English Summary (Experimental)"],
            label="Summarization Type",
            value="Turkish Text -> Turkish Summary"
        ),
        gr.Slider(
            minimum=10, 
            maximum=100, 
            value=10, # Default value for natural, short summaries
            step=5, 
            label="Minimum Summary Length (Tokens)",
            info="Increase this value to force the model to generate longer summaries."
        )
    ],
    outputs=gr.Textbox(lines=5, label="Generated Summary"),
    title="Multilingual Text Summarization Model",
    description="""This demo generates headline-style summaries for Turkish or English texts.
    The model was fine-tuned on 30,000 Turkish-English summary pairs using the LoRA technique, 
    based on the `google/mt5-small` model.""",
    
    # The values in the examples list are now 100% consistent with the Radio button choices.
    examples=[
        ["Güneş enerjisi kullanımı son yıllarda hızla artıyor. Evler ve iş yerleri çatılarında güneş panelleri kurarak elektrik üretiyor. Bu sayede enerji faturaları düşüyor ve çevreye zarar vermeyen temiz enerji kullanılıyor. Uzmanlar, güneş enerjisinin gelecekte elektrik ihtiyacının büyük kısmını karşılayabileceğini ve fosil yakıtlara olan bağımlılığı azaltacağını söylüyor.", "Turkish Text -> Turkish Summary", 10],
        ["Electric cars are becoming increasingly popular worldwide. People choose electric vehicles because they produce less pollution and reduce fuel costs. Governments are promoting them by offering incentives and building more charging stations. Experts believe that electric cars could eventually replace most gasoline-powered vehicles, significantly decreasing greenhouse gas emissions and contributing to a cleaner environment.", "English Text -> English Summary", 10]
    ],
    theme="gradio/soft",
    allow_flagging="never"
)

if __name__ == "__main__":
    iface.launch()