app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from peft import PeftModel
import torch

# Load model and tokenizer
print("Loading model...")
tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-base")
base_model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-base")

# Replace with your actual model repository name
model = PeftModel.from_pretrained(base_model, "noviciusss/flan-t5-base-samsum")
model.eval()

print("Model loaded successfully!")

def summarize_dialogue(dialogue, max_length=128, num_beams=4):
    """Summarizes a dialogue using the fine-tuned FLAN-T5 model."""
    if not dialogue.strip():
        return "Please enter a dialogue to summarize."
    
    # Prepare input
    input_text = "summarize: " + dialogue
    inputs = tokenizer(
        input_text,
        return_tensors="pt",
        max_length=512,
        truncation=True
    )
    
    # Generate summary
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_length=int(max_length),
            num_beams=int(num_beams),
            early_stopping=True,
            no_repeat_ngram_size=3
        )
    
    # Decode and return
    summary = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return summary

# Example dialogues with ALL input values
examples = [
    ["""Hannah: Hey, did you see the game last night?
Adam: Yeah! That last-minute goal was incredible!
Hannah: I know right? Best match of the season so far.
Adam: Totally agree. We should watch the next game together.
Hannah: For sure! I'll bring snacks.""", 128, 4],
    
    ["""Mike: Can you pick up some groceries on your way home?
Sarah: Sure, what do we need?
Mike: Milk, bread, and eggs. Oh, and some coffee.
Sarah: Got it. Anything else?
Mike: That's all. Thanks!""", 128, 4],
    
    ["""Emma: What time is the meeting tomorrow?
John: It's at 2 PM in the conference room.
Emma: Great, I'll prepare the presentation.
John: Perfect. See you then!""", 96, 4]
]

# Create Gradio interface
demo = gr.Interface(
    fn=summarize_dialogue,
    inputs=[
        gr.Textbox(
            label="Dialogue",
            placeholder="Enter a conversation to summarize...",
            lines=10
        ),
        gr.Slider(
            minimum=32,
            maximum=256,
            value=128,
            step=16,
            label="Max Summary Length"
        ),
        gr.Slider(
            minimum=1,
            maximum=8,
            value=4,
            step=1,
            label="Number of Beams"
        )
    ],
    outputs=gr.Textbox(label="Summary", lines=3),
    examples=examples,
    cache_examples=False,  # Disable caching to avoid the error
    title="🗨️ Dialogue Summarizer",
    description="""
    This app summarizes conversational dialogues using a **FLAN-T5 base** model fine-tuned with **LoRA** on the SAMSum dataset.
    
    **Performance Metrics:**
    - ROUGE-1: 49.01 | ROUGE-2: 25.06 | ROUGE-L: 40.97
    - BERTScore F1: 72.25 | METEOR: 42.51
    
    Enter a dialogue or try one of the examples below!
    """,
    article="""
    <div style='text-align: center; margin-top: 20px;'>
        <p>Model: <a href='https://huggingface.co/YOUR_USERNAME/SamSum' target='_blank'>noviciusss/flan-t5-base-samsum</a></p>
        <p>Built with 🤗 Transformers, PEFT, and Gradio</p>
    </div>
    """,
    theme=gr.themes.Soft(),
    allow_flagging="never"
)

# Launch the app
if __name__ == "__main__":
    demo.launch()