Update app.py

app.py

@@ -1,10 +1,10 @@
 """
 SchoolSpirit AI – Granite‑3.3‑2B chatbot Space
 ----------------------------------------------
-• Uses IBM Granite‑3.3‑2B‑Instruct (Apache‑2).
-• Keeps last MAX_TURNS exchanges.
-• “Clear chat” button resets context.
-• Robust error handling & logging.
+• IBM Granite‑3.3‑2B‑Instruct (Apache‑2), runs in HF CPU Space.
+• Keeps last MAX_TURNS exchanges to fit context.
+• “Clear Chat” button resets conversation.
+• Extensive error‑handling: model‑load, inference, bad input.
 """
 
 import re
@@ -16,27 +16,29 @@ from transformers import (
 )
 from transformers.utils import logging as hf_logging
 
-# ─────────────── Config ────────────────────────────────────────────────────
+# ────────── Configuration ───────────────────────────────────────────────────
 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
+MAX_TURNS    = 6          # history turns to keep
+MAX_TOKENS   = 200        # response length
+MAX_INPUT_CH = 400        # user message length guard
 
 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."
+    "say so and suggest contacting a human. Never request personal data."
 )
 
-# ─────────────── Model Load ────────────────────────────────────────────────
+# ────────── Model loading with fail‑safe ────────────────────────────────────
 try:
     tok = AutoTokenizer.from_pretrained(MODEL_ID)
     model = AutoModelForCausalLM.from_pretrained(
-        MODEL_ID, device_map="auto", torch_dtype="auto"
+        MODEL_ID,
+        device_map="auto",
+        torch_dtype="auto",
     )
     generator = pipeline(
         "text-generation",
@@ -52,18 +54,35 @@ except Exception as exc:  # noqa: BLE001
     generator = None
     LOG.error(MODEL_ERR)
 
-# ─────────────── Helpers ───────────────────────────────────────────────────
+# ────────── Helper utilities ────────────────────────────────────────────────
 def truncate(hist):
+    """Return last MAX_TURNS (user,bot) tuples."""
     return hist[-MAX_TURNS:] if len(hist) > MAX_TURNS else hist
 
 
 def clean(text: str) -> str:
+    """Normalize whitespace and guarantee non‑empty."""
     return re.sub(r"\s+", " ", text.strip()) or "…"
 
-# ─────────────── Chat Callback ─────────────────────────────────────────────
+
+def safe_generate(prompt: str) -> str:
+    """Call model.generate, catch & log any error, always return a string."""
+    try:
+        completion = generator(prompt)[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 in a moment."
+        )
+    return reply
+
+# ────────── Chat callback ───────────────────────────────────────────────────
 def chat(history, user_msg):
-    history = list(history)
+    history = list(history)  # guaranteed list of tuples
 
+    # Fatal start‑up failure
     if MODEL_ERR:
         history.append((user_msg, MODEL_ERR))
         return history, ""
@@ -73,35 +92,33 @@ def chat(history, 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."))
+        history.append(
+            (user_msg, f"Message too long (>{MAX_INPUT_CH} chars).")
+        )
         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)[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."
+    reply = safe_generate(prompt)
 
     history.append((user_msg, reply))
     return history, ""
 
-# ─────────────── Clear Chat ────────────────────────────────────────────────
+# ────────── Clear chat callback ─────────────────────────────────────────────
 def clear_chat():
     return [], ""
 
-# ─────────────── UI Launch ────────────────────────────────────────────────
+# ────────── UI definition ───────────────────────────────────────────────────
 with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as demo:
     gr.Markdown("# SchoolSpirit AI Chat")
-    chatbot   = gr.Chatbot(type="tuple")  # legacy tuple format
+    chatbot   = gr.Chatbot(type="tuples")
     msg_box   = gr.Textbox(placeholder="Ask me anything about SchoolSpirit AI…")
     send_btn  = gr.Button("Send")
     clear_btn = gr.Button("Clear Chat", variant="secondary")