# app.py — Gradio App for Text Summarization
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

MODEL_ID = "samandar1105/text-summarizer"
PREFIX = "summarize: "
MAX_INPUT_LENGTH = 512

# ... (paste the rest of the app.py code exactly as shown in Phase 5) ...

# app.py — Gradio App for Text Summarization
import gradio as gr
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

# ============================================================
# CONFIGURATION — Update this to your model!
# ============================================================
MODEL_ID = "samandar1105/text-summarizer"   # ← CHANGE THIS
PREFIX = "summarize: "                                    # T5 needs this prefix
MAX_INPUT_LENGTH = 512

EXAMPLES = [
    ["""NASA's Perseverance rover has collected its most compelling sample yet in the
search for ancient life on Mars, scientists announced Wednesday. The rock sample,
nicknamed "Cheyava Falls," contains chemical signatures and structures that could
be consistent with microbial life billions of years ago, though researchers caution
that non-biological explanations have not been ruled out. The sample will eventually
be returned to Earth for detailed laboratory analysis as part of the Mars Sample
Return mission, currently planned for the early 2030s."""],
    ["""The Federal Reserve held interest rates steady at its policy meeting on
Wednesday, extending a pause that has lasted several months as officials continue
to monitor inflation data. In a statement, the central bank said recent indicators
suggest the economy is expanding at a solid pace, while inflation has eased over the
past year but remains somewhat elevated relative to the 2% target. Markets had
widely expected the decision, and futures pricing suggests investors are looking to
the next meeting for signs of a potential rate cut."""],
]

# We load the tokenizer/model directly and call .generate() ourselves instead of
# using pipeline("summarization", ...). Some environments have a broken or
# mismatched pipeline task registry that raises "KeyError: Unknown task
# summarization" even when transformers is otherwise installed correctly — this
# approach sidesteps that entirely and also gives us direct control over beam
# search and repetition settings.
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Loading model: {MODEL_ID} on {DEVICE}")
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID).to(DEVICE)
print("Model loaded successfully!")


def summarize_text(text: str, max_len: int, min_len: int):
    if not text or text.strip() == "":
        return "⚠️ Please paste some text to summarize."
    if len(text.strip().split()) < 15:
        return "⚠️ Please enter a longer passage (at least ~15 words) for a meaningful summary."

    inputs = tokenizer(
        PREFIX + text.strip(),
        return_tensors="pt",
        truncation=True,
        max_length=MAX_INPUT_LENGTH,
    ).to(DEVICE)

    output_ids = model.generate(
        **inputs,
        max_length=int(max_len),
        min_length=int(min_len),
        num_beams=4,
        no_repeat_ngram_size=3,
    )
    return tokenizer.decode(output_ids[0], skip_special_tokens=True)


with gr.Blocks(title="📝 AI Text Summarizer", theme=gr.themes.Soft()) as demo:
    gr.Markdown("""
    # 📝 AI Text Summarizer
    Paste any article, report, or long passage below and get a concise AI-generated summary.
    Built with `t5-small` fine-tuned on the CNN/DailyMail dataset.
    """)

    with gr.Row():
        with gr.Column(scale=2):
            input_text = gr.Textbox(
                label="📄 Text to summarize",
                placeholder="Paste your article or long text here...",
                lines=12,
            )
            with gr.Row():
                max_len_slider = gr.Slider(30, 200, value=120, step=10, label="Max summary length")
                min_len_slider = gr.Slider(5, 60, value=20, step=5, label="Min summary length")
            with gr.Row():
                submit_btn = gr.Button("✨ Summarize", variant="primary", scale=2)
                clear_btn = gr.ClearButton([input_text], scale=1)

        with gr.Column(scale=2):
            output_text = gr.Textbox(label="📌 Summary", lines=8)

    gr.Examples(examples=EXAMPLES, inputs=input_text, label="📌 Try an example")

    gr.Markdown("---\n**Model:** `t5-small` fine-tuned on CNN/DailyMail")

    submit_btn.click(
        fn=summarize_text,
        inputs=[input_text, max_len_slider, min_len_slider],
        outputs=[output_text],
    )
    input_text.submit(
        fn=summarize_text,
        inputs=[input_text, max_len_slider, min_len_slider],
        outputs=[output_text],
    )

if __name__ == "__main__":
    demo.launch(server_port=7860)