app.py

"""
SchoolSpirit AI – Granite‑3.3‑2B chatbot Space
----------------------------------------------
• Uses IBM Granite‑3.3‑2B‑Instruct (public, no access token).
• Fits HF CPU Space (2‑B params, bfloat16).
• Keeps last MAX_TURNS exchanges.
• “Clear chat” button resets context.
• Robust error handling & logging.
"""

import re
import gradio as gr
from transformers import (
    AutoTokenizer,
    AutoModelForCausalLM,
    pipeline,
)
from transformers.utils import logging as hf_logging

# ────────────────────────── Config ──────────────────────────────────────────
hf_logging.set_verbosity_error()
LOG = hf_logging.get_logger("SchoolSpirit")

MODEL_ID     = "ibm-granite/granite-3.3-2b-instruct"
MAX_TURNS    = 6
MAX_TOKENS   = 200
MAX_INPUT_CH = 400

SYSTEM_MSG = (
    "You are SchoolSpirit AI, the upbeat digital mascot for a company that "
    "offers on‑prem AI chat mascots, fine‑tuning services, and turnkey GPU "
    "hardware for schools. Answer concisely and age‑appropriately. If unsure, "
    "say so and suggest contacting a human. Do not ask for personal data."
)

# ────────────────────────── Model Load ──────────────────────────────────────
try:
    tok = AutoTokenizer.from_pretrained(MODEL_ID)
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_ID,
        device_map="auto",
        torch_dtype="auto",          # bfloat16/float16 under the hood
    )
    generator = pipeline(
        "text-generation",
        model=model,
        tokenizer=tok,
        max_new_tokens=MAX_TOKENS,
        do_sample=True,
        temperature=0.7,
    )
    MODEL_ERR = None
except Exception as exc:  # noqa: BLE001
    MODEL_ERR = f"Model load error: {exc}"
    generator = None
    LOG.error(MODEL_ERR)

# ────────────────────────── Helpers ────────────────────────────────────────
def truncate(hist):
    """Return last MAX_TURNS (u,a) pairs."""
    return hist[-MAX_TURNS:] if len(hist) > MAX_TURNS else hist


def clean(text: str) -> str:
    """Collapse whitespace; never return empty string."""
    out = re.sub(r"\s+", " ", text.strip())
    return out or "…"

# ────────────────────────── Chat Callback ───────────────────────────────────
def chat(history, user_msg):
    history = list(history)  # Gradio ensures list of tuples

    if MODEL_ERR:
        history.append((user_msg, MODEL_ERR))
        return history, ""

    user_msg = clean(user_msg or "")
    if not user_msg:
        history.append(("", "Please enter a message."))
        return history, ""
    if len(user_msg) > MAX_INPUT_CH:
        history.append((user_msg, "That message is too long."))
        return history, ""

    history = truncate(history)

    # Build prompt
    prompt_lines = [SYSTEM_MSG]
    for u, a in history:
        prompt_lines += [f"User: {u}", f"AI: {a}"]
    prompt_lines += [f"User: {user_msg}", "AI:"]
    prompt = "\n".join(prompt_lines)

    try:
        completion = generator(prompt, truncate=4096)[0]["generated_text"]
        reply = clean(completion.split("AI:", 1)[-1])
    except Exception as err:  # noqa: BLE001
        LOG.error(f"Inference error: {err}")
        reply = "Sorry—I'm having trouble right now. Please try again shortly."

    history.append((user_msg, reply))
    return history, ""

# ────────────────────────── Clear Chat ──────────────────────────────────────
def clear_chat():
    return [], ""

# ────────────────────────── UI Launch ───────────────────────────────────────
with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as demo:
    gr.Markdown("# SchoolSpirit AI Chat")
    chatbot  = gr.Chatbot()
    msg_box  = gr.Textbox(placeholder="Ask me anything about SchoolSpirit AI…")
    send_btn = gr.Button("Send")
    clear_btn = gr.Button("Clear Chat", variant="secondary")

    send_btn.click(chat, [chatbot, msg_box], [chatbot, msg_box])
    msg_box.submit(chat, [chatbot, msg_box], [chatbot, msg_box])
    clear_btn.click(clear_chat, outputs=[chatbot, msg_box])

demo.launch()