Update app.py

app.py

@@ -1,27 +1,38 @@
 """
-ZEN-Bot Lite  –  Key-free chatbot for HF Spaces (free CPU)
+ZEN-Bot (stable build) – Hugging Face Space, no API keys
 Skills
-  • Natural chat
+  • Friendly chat
   • /math        – safe calculator
-  • /summarize   – 2-sentence abstract
-  • /translate_es – English → Spanish
-Only one open-weights model (facebook/blenderbot-400M-distill).
+  • /summarize   – 2-sentence TL;DR
+  • /translate_es – English ➜ Spanish
+Model: microsoft/DialoGPT-medium (~762 MB weights, runs on free CPU)
 """
 
-import ast, math, re, gc
+import ast, math, re, gc, torch, traceback
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
 # ---------------------------------------------------------------------
-# 1 · Model small enough for free CPU
+# 0 · Small helpers
 # ---------------------------------------------------------------------
-MODEL_NAME = "facebook/blenderbot-400M-distill"   # ~720 MB on disk, loads in ≈3 GB RAM
+def log_err(e: Exception) -> str:
+    print("=== ZEN-Bot ERROR ===")
+    traceback.print_exc()
+    print("=====================")
+    return f"⚠️ {type(e).__name__}: {e}"
+
+# ---------------------------------------------------------------------
+# 1 · Load model & tokenizer (fits free CPU tier)
+# ---------------------------------------------------------------------
+MODEL_NAME = "microsoft/DialoGPT-medium"
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model     = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
+model     = AutoModelForCausalLM.from_pretrained(MODEL_NAME)
+model.eval()                      # we’re only generating, not training
+torch.set_grad_enabled(False)     # global no-grad just in case
 
 # ---------------------------------------------------------------------
-# 2 · Safe /math evaluator  (same as before)
+# 2 · Safe /math evaluator
 # ---------------------------------------------------------------------
 _ALLOWED = {k: getattr(math, k) for k in dir(math) if not k.startswith("__")}
 _ALLOWED.update({"abs": abs, "round": round})
@@ -31,101 +42,112 @@ def safe_math(expr: str) -> str:
         node = ast.parse(expr, mode="eval")
 
         def ok(n):
-            if isinstance(n, ast.Num):
+            if isinstance(n, ast.Num):            # numbers
                 return True
-            if isinstance(n, ast.UnaryOp):
+            if isinstance(n, ast.UnaryOp):        # -x
                 return ok(n.operand)
-            if isinstance(n, ast.BinOp):
+            if isinstance(n, ast.BinOp):          # x + y
                 return ok(n.left) and ok(n.right)
-            if isinstance(n, ast.Call):
-                return (isinstance(n.func, ast.Name) and
-                        n.func.id in _ALLOWED and
-                        all(ok(a) for a in n.args))
+            if isinstance(n, ast.Call):           # sin(x)
+                return (isinstance(n.func, ast.Name)
+                        and n.func.id in _ALLOWED
+                        and all(ok(a) for a in n.args))
             return False
 
         if not ok(node.body):
-            return "⛔️ Expression not allowed"
+            return "⛔️ Only basic math functions are allowed."
         return str(eval(compile(node, "<expr>", "eval"),
                         {"__builtins__": {}}, _ALLOWED))
     except Exception as e:
-        return f"⚠️ Error: {e}"
+        return log_err(e)
 
 # ---------------------------------------------------------------------
-# 3 · Generation helper
+# 3 · LLM generation helper
 # ---------------------------------------------------------------------
 GEN_KW = dict(
-    max_new_tokens = 128,
-    do_sample      = True,
-    top_p          = 0.92,
-    temperature    = 0.7,
+    max_new_tokens = 64,          # fast & HF 30-sec-safe
+    do_sample      = False,       # deterministic, fewer nonsense tokens
+    pad_token_id   = tokenizer.eos_token_id,
 )
 
 def llm(prompt: str) -> str:
-    input_ids  = tokenizer(prompt, return_tensors="pt").input_ids
-    output_ids = model.generate(**GEN_KW, input_ids=input_ids)
-    reply_ids  = output_ids[0, input_ids.shape[-1]:]
-    return tokenizer.decode(reply_ids, skip_special_tokens=True).strip()
+    try:
+        input_ids  = tokenizer(prompt, return_tensors="pt").input_ids
+        with torch.no_grad():
+            output_ids = model.generate(**GEN_KW, input_ids=input_ids)
+        reply_ids  = output_ids[0, input_ids.shape[-1]:]
+        return tokenizer.decode(reply_ids, skip_special_tokens=True).strip()
+    except Exception as e:
+        return log_err(e)
 
 # ---------------------------------------------------------------------
-# 4 · Regex router + lightweight name memory (stored inside history)
+# 4 · Command router + tiny name memory (history-local)
 # ---------------------------------------------------------------------
-CMD = re.compile(r"^/(math|summarize|translate_es)\s+(.+)", re.S | re.I)
-
-def reply(message: str, history: list[list[str, str]]) -> str:
-    # 4.1 · Command handling first
-    m = CMD.match(message.strip())
-    if m:
-        cmd, payload = m.group(1).lower(), m.group(2).strip()
-        if cmd == "math":
-            return safe_math(payload)
-
-        if cmd == "summarize":
-            prompt = ( "Summarize in two concise sentences:\n\n"
-                       f"{payload}\n\nSummary:" )
-            return llm(prompt)
-
-        if cmd == "translate_es":
-            prompt = ( "Translate the following text from English to Spanish "
-                       "(natural, not literal):\n\n"
-                       f"{payload}\n\nSpanish:" )
-            return llm(prompt)
-
-    # 4.2 · Grab user’s name if volunteered
-    name_match = re.search(r"\bmy name is (\w+)", message, re.I)
-    name = name_match.group(1).capitalize() if name_match else None
-
-    # 4.3 · Build dialogue prompt
-    system = "You are ZEN-Bot, a concise, friendly tutor for young AI pioneers."
-    if name:
-        system += f" The user's name is {name}."
-    prompt = system + "\n\n"
-    for u, b in history:
-        prompt += f"User: {u}\nAssistant: {b}\n"
-    prompt += f"User: {message}\nAssistant:"
-
-    return llm(prompt)
+COMMAND = re.compile(r"^/(math|summarize|translate_es)\s+(.+)", re.S | re.I)
+
+def respond(message: str, history: list[list[str, str]]) -> str:
+    try:
+        # 4.1 · Commands
+        m = COMMAND.match(message.strip())
+        if m:
+            cmd, payload = m.group(1).lower(), m.group(2).strip()
+            if cmd == "math":
+                return safe_math(payload)
+
+            if cmd == "summarize":
+                prompt = (
+                    "Summarize the following text in two concise sentences:\n\n"
+                    f"{payload}\n\nSummary:"
+                )
+                return llm(prompt)
+
+            if cmd == "translate_es":
+                prompt = (
+                    "Translate the following text from English to Spanish "
+                    "(natural, not literal):\n\n"
+                    f"{payload}\n\nSpanish:"
+                )
+                return llm(prompt)
+
+        # 4.2 · Capture user’s name (per session)
+        name_match = re.search(r"\bmy name is (\w+)", message, re.I)
+        name_line  = ""
+        if name_match:
+            name_line = f" The user's name is {name_match.group(1).capitalize()}."
+
+        # 4.3 · Build chat prompt
+        system = "You are ZEN-Bot, a concise, friendly tutor for young AI pioneers." + name_line
+        prompt = system + "\n\n"
+        for u, b in history[-6:]:                 # keep last 6 turns for speed
+            prompt += f"User: {u}\nAssistant: {b}\n"
+        prompt += f"User: {message}\nAssistant:"
+
+        return llm(prompt)
+
+    except Exception as e:
+        return log_err(e)
 
 # ---------------------------------------------------------------------
 # 5 · Gradio UI
 # ---------------------------------------------------------------------
 demo = gr.ChatInterface(
-    fn          = reply,
-    title       = "🔧 ZEN-Bot Lite (Key-Free)",
+    fn          = respond,
+    title       = "✨ ZEN-Bot (Key-Free Edition)",
     description = (
-        "**Try me**\n"
-        "• normal chat\n"
-        "• `/math 3*(sin(0.5)+1)`\n"
-        "• `/summarize Industrial Revolution text...`\n"
-        "• `/translate_es Good morning!`"
+        "**What I can do**\n"
+        "• Just chat normally\n"
+        "• `/math 2**5 / (sin(0.5)+1)`\n"
+        "• `/summarize <any text>`\n"
+        "• `/translate_es Hello, how are you?`"
     ),
-    examples    = [
-        "Hi, my name is Jordan!",
-        "/math 2**6 / 3",
-        "/summarize The water cycle begins when...",
-        "/translate_es We are building awesome AI projects.",
-    ],
     theme       = "soft",
     fill_height = True,
+    examples    = [
+        "Hi, my name is Sam!",
+        "/math log(10) + sqrt(16)",
+        "/summarize The Internet began as a US-funded research project in 1969…",
+        "/translate_es Artificial intelligence is transforming education.",
+    ],
 )
 
 if __name__ == "__main__":