Update app.py

app.py

@@ -8,56 +8,45 @@ Skills
   • /translate_es <text>   — English → Spanish
   • /ascii <text>          — FIGlet ASCII art
   • Any question ending with “?” → live Wikipedia answer
-     ↳ remembers the last topic, so “Who discovered it?” now works.
-Model: facebook/blenderbot-400M-distill  (~720 MB, public)
+     ↳ remembers last topic, so “Who discovered it?” works.
 
-Copy-paste this file into your Space, keep the same requirements.txt, commit, done.
+Model: facebook/blenderbot-400M-distill  (public, ~720 MB)
 """
 
 import ast, math, re, gc, traceback, torch, wikipedia, pyfiglet
 import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
 
-# ────────────────────────────────────────────────────────────────
-# 0 · helpers
-# ────────────────────────────────────────────────────────────────
+# ───────────────────────── helpers ──────────────────────────
 def log_err(e: Exception) -> str:
-    print("=== ZEN-Bot ERROR =====================")
+    print("=== ZEN-Bot ERROR ===")
     traceback.print_exc()
-    print("=======================================\n")
+    print("=====================\n")
     return f"⚠️ {type(e).__name__}: {e}"
 
-# ────────────────────────────────────────────────────────────────
-# 1 · model (public, small enough for free CPU/basic)
-# ────────────────────────────────────────────────────────────────
-MODEL = "facebook/blenderbot-400M-distill"
+# ─────────────────────── model loading ──────────────────────
+MODEL_ID = "facebook/blenderbot-400M-distill"   # always public
 
-tokenizer = AutoTokenizer.from_pretrained(MODEL)
-model     = AutoModelForSeq2SeqLM.from_pretrained(MODEL)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model     = AutoModelForSeq2SeqLM.from_pretrained(MODEL_ID)
 model.eval();  torch.set_grad_enabled(False)
 
-GEN_ARGS = dict(max_new_tokens=64, do_sample=False)   # greedy, fast
+GEN_KW = dict(max_new_tokens=64, do_sample=False)   # deterministic → faster
 
 def llm(prompt: str) -> str:
-    """Generate with BlenderBot-distill, catching any runtime errors."""
     try:
-        inputs = tokenizer(
-            prompt, return_tensors="pt",
-            truncation=True, max_length=1024
-        ).input_ids
+        ids_in = tokenizer(prompt, return_tensors="pt",
+                           truncation=True, max_length=1024).input_ids
         with torch.no_grad():
-            out = model.generate(inputs, **GEN_ARGS)
-        answer_ids = out[0, inputs.shape[-1]:]
-        return tokenizer.decode(answer_ids,
-                                skip_special_tokens=True).strip()
+            ids_out = model.generate(ids_in, **GEN_KW)
+        reply_ids = ids_out[0, ids_in.shape[-1]:]
+        return tokenizer.decode(reply_ids, skip_special_tokens=True).strip()
     except Exception as e:
         return log_err(e)
 
-# ────────────────────────────────────────────────────────────────
-# 2 · safe /math
-# ────────────────────────────────────────────────────────────────
-_ALLOWED = {k: getattr(math, k) for k in dir(math) if not k.startswith("_")}
-_ALLOWED.update({"abs": abs, "round": round})
+# ──────────────────────── /math safe-eval ───────────────────
+_MATH = {k: getattr(math, k) for k in dir(math) if not k.startswith("_")}
+_MATH.update({"abs": abs, "round": round})
 
 def safe_math(expr: str) -> str:
     try:
@@ -65,102 +54,88 @@ def safe_math(expr: str) -> str:
 
         def ok(n):
             match n:
-                case ast.Num():                    return True
-                case ast.UnaryOp():                return ok(n.operand)
-                case ast.BinOp():                  return ok(n.left) and ok(n.right)
+                case ast.Num():           return True
+                case ast.UnaryOp():       return ok(n.operand)
+                case ast.BinOp():         return ok(n.left) and ok(n.right)
                 case ast.Call():
                     return (isinstance(n.func, ast.Name)
-                            and n.func.id in _ALLOWED
+                            and n.func.id in _MATH
                             and all(ok(a) for a in n.args))
-                case _:                            return False
+                case _:                   return False
 
         if not ok(node.body):
-            return "⛔️ Only basic maths & trig/log functions allowed."
+            return "⛔️ Only basic math / trig / log functions allowed."
         return str(eval(compile(node, "<expr>", "eval"),
-                        {"__builtins__": {}}, _ALLOWED))
+                        {"__builtins__": {}}, _MATH))
     except Exception as e:
         return log_err(e)
 
-# ────────────────────────────────────────────────────────────────
-# 3 · Wikipedia Q&A (with last-topic memory)
-# ────────────────────────────────────────────────────────────────
-last_topic: dict[str, str] = {}   # {session_hash: "Penicillin"}
+# ──────────────────────── Wikipedia Q&A ─────────────────────
+last_topic: str | None = None     # shared across the single free CPU Space
 
-def wiki_answer(q: str, session_id: str) -> str | None:
-    """Return a 3-sentence Wikipedia summary or None if not found."""
+def wiki_answer(q: str) -> str | None:
+    """3-sentence Wikipedia answer — returns None if not found."""
+    global last_topic
     try:
-        # If the question uses 'it', swap with last remembered topic
         query = q.strip()
-        if last_topic.get(session_id) and re.search(r"\bit\b", query, re.I):
-            query = re.sub(r"\bit\b", last_topic[session_id], query, flags=re.I)
+        if last_topic and re.search(r"\bit\b", query, re.I):
+            query = re.sub(r"\bit\b", last_topic, query, flags=re.I)
 
         wikipedia.set_lang("en")
         page = wikipedia.page(query, auto_suggest=True, redirect=True)
-        last_topic[session_id] = page.title     # remember for pronouns
-        summary = wikipedia.summary(
-            page.title, sentences=3, auto_suggest=False
-        )
+        last_topic = page.title               # remember for next turn
+        summary = wikipedia.summary(page.title, sentences=3, auto_suggest=False)
         return f"**{page.title}** — {summary}"
     except (wikipedia.DisambiguationError, wikipedia.PageError):
-        return None           # let LLM try instead
+        return None
     except Exception as e:
         return log_err(e)
 
-# ────────────────────────────────────────────────────────────────
-# 4 · ASCII art
-# ────────────────────────────────────────────────────────────────
-def ascii_art(txt: str) -> str:
+# ───────────────────────── ASCII art ─────────────────────────
+def ascii_art(text: str) -> str:
     try:
-        return pyfiglet.figlet_format(txt, width=120)
+        return pyfiglet.figlet_format(text, width=120)
     except Exception as e:
         return log_err(e)
 
-# ────────────────────────────────────────────────────────────────
-# 5 · router ↔ ChatInterface callback
-# ────────────────────────────────────────────────────────────────
+# ────────────────── main router / callback ───────────────────
 CMD = re.compile(r"^/(math|summarize|translate_es|ascii)\s+(.+)", re.S | re.I)
 
-def respond(message: str,
-            history: list[list[str, str]],
-            request: gr.Request) -> str:
-    session_id = request.session_hash or "anon"
-
-    # 5-A · Command shortcuts
-    if (m := CMD.match(message.strip())):
+def respond(msg: str, history: list[list[str, str]]) -> str:
+    # A · commands
+    if (m := CMD.match(msg.strip())):
         cmd, body = m.group(1).lower(), m.group(2).strip()
         if cmd == "math":         return safe_math(body)
         if cmd == "ascii":        return ascii_art(body)
-        if cmd == "summarize":    return llm(f"Summarise in two concise sentences:\n\n{body}\n\nSummary:")
+        if cmd == "summarize":    return llm(f"Summarize in two concise sentences:\n\n{body}\n\nSummary:")
         if cmd == "translate_es": return llm(f"Translate into Spanish (natural):\n\n{body}\n\nSpanish:")
 
-    # 5-B · Wikipedia factual Q&A
-    if message.strip().endswith("?") and len(message.split()) > 2:
-        wiki = wiki_answer(message, session_id)
+    # B · live Wikipedia for factual Qs
+    if msg.endswith("?") and len(msg.split()) > 2:
+        wiki = wiki_answer(msg)
         if wiki:  return wiki
 
-    # 5-C · Normal chat  (keep last 6 turns for speed)
-    prompt = ("You are ZEN-Bot, a concise, friendly tutor for young AI pioneers.\n\n")
+    # C · normal chat (keep last 6 turns for speed)
+    prompt = "You are ZEN-Bot, a concise, friendly tutor for young AI pioneers.\n\n"
     for u, b in history[-6:]:
         prompt += f"User: {u}\nAssistant: {b}\n"
-    prompt += f"User: {message}\nAssistant:"
+    prompt += f"User: {msg}\nAssistant:"
     return llm(prompt)
 
-# ────────────────────────────────────────────────────────────────
-# 6 · Gradio UI
-# ────────────────────────────────────────────────────────────────
+# ────────────────────── Gradio Chat UI ───────────────────────
 demo = gr.ChatInterface(
-    fn          = respond,
-    title       = "🚀 ZEN-Bot Ultimate (Key-Free)",
-    description = (
+    fn               = respond,
+    title            = "🚀 ZEN-Bot Ultimate (Key-Free)",
+    description      = (
         "**Commands**  \n"
-        "• Normal chat  \n"
+        "• normal chat  \n"
         "• `/math 2**5 / (sin(0.5)+1)`  \n"
         "• `/summarize <text>`  \n"
         "• `/translate_es Hello!`  \n"
         "• `/ascii ZEN`  \n"
-        "• Ask factual questions ending with `?` (remembers topic for pronouns)"
+        "• Ask factual questions ending with `?` (remembers topic for 'it')"
     ),
-    examples    = [
+    examples         = [
         "Who discovered penicillin?",
         "/ascii AI ROCKS",
         "/math sqrt(144)+log(100,10)",
@@ -168,8 +143,9 @@ demo = gr.ChatInterface(
         "/translate_es Good evening, friends!",
         "Who discovered it?",
     ],
-    fill_height = True,
-    theme       = "soft",
+    cache_examples   = False,          # ← avoids the startup pre-run
+    theme            = "soft",
+    fill_height      = True,
 )
 
 if __name__ == "__main__":