app.py

import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from huggingface_hub import login
import os

# -----------------------------
# 1. Login using HF token (from environment variable)
# -----------------------------
hf_token = os.environ.get("HF_TOKEN")
if hf_token:
    login(token=hf_token)
else:
    print("⚠️ Warning: HF_TOKEN not set. Add it in the Space settings.")

# -----------------------------
# 2. Load Model
# -----------------------------
model_id = "ibm-granite/granite-3.3-2b-instruct"

tokenizer = AutoTokenizer.from_pretrained(model_id, use_auth_token=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
    use_auth_token=True
)

# -----------------------------
# 3. Functions
# -----------------------------
def classify_sentiment(text):

    prompt = f"""
Classify the sentiment of the following review as one word: Positive, Negative, or Neutral.
Review: "{text}"
Only respond with the sentiment label.
"""

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    output = model.generate(
        **inputs,
        max_new_tokens=5,
        temperature=0.1,
        repetition_penalty=1.5
    )

    result = tokenizer.decode(output[0], skip_special_tokens=True)

    final = result.split()[-1].strip()

    return final


def summarize_text(text):

    prompt = f"""
Summarize the following text in 3–5 bullet points.
Make the summary short, simple, and avoid repeating sentences.
Text: {text}
Summary:
"""

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    output = model.generate(
        **inputs,
        max_new_tokens=150,
        temperature=0.7,
        top_p=0.9,
        repetition_penalty=2.0,
    )

    result = tokenizer.decode(output[0], skip_special_tokens=True)

    summary = result.split("Summary:")[-1].strip()

    return summary



# -----------------------------
# 4. Gradio Interface
# -----------------------------
with gr.Blocks() as demo:

    gr.Markdown("## 🟦 Granite 3.3-2B — Sentiment + Summarization Demo")

    with gr.Tab("Sentiment Classification"):
        inp = gr.Textbox(label="Enter Review", lines=4)
        out = gr.Textbox(label="Predicted Sentiment", lines=1)
        btn = gr.Button("Classify")
        btn.click(classify_sentiment, inp, out)

    with gr.Tab("Summarization"):
        inp2 = gr.Textbox(label="Enter Text to Summarize", lines=8)
        out2 = gr.Textbox(label="Summary", lines=8)
        btn2 = gr.Button("Summarize")
        btn2.click(summarize_text, inp2, out2)

demo.launch()