Update app.py

app.py

@@ -70,54 +70,51 @@ smol_agent = CodeAgent(
     max_steps=5
 )
 
-# ============ 2. LLAMAINDEX SETUP (Fixed) ============
+# ============ 2. LLAMAINDEX SETUP (STABLE VERSION) ============
 
-# FIX 1: Define real functions with docstrings so the Agent knows what they do
+# FIX 1: Use 32B for LlamaIndex too. 7B is too weak for "ReAct" reasoning.
+# FIX 2: Use task="text-generation" (Stable for HF Inference API)
+llama_llm = HuggingFaceInferenceAPI(
+    model_name="Qwen/Qwen2.5-Coder-32B-Instruct", 
+    token=HF_TOKEN,
+    task="text-generation",
+    timeout=60
+)
+
+# FIX 3: ReActAgent needs very clear instructions to stay in "Agent Mode"
+# We define tools with explicit Type Hints and Docstrings
 def get_tokyo_time() -> str:
-    """
-    Returns the current time in Tokyo, Japan.
-    Use this tool when the user asks for the time in Tokyo.
-    """
-    return f"Tokyo: {datetime.datetime.now(pytz.timezone('Asia/Tokyo'))}"
+    """Returns the current date and time in Tokyo, Japan."""
+    tz = pytz.timezone('Asia/Tokyo')
+    return f"Tokyo Time: {datetime.datetime.now(tz).strftime('%Y-%m-%d %H:%M:%S')}"
 
 def multiply(a: float, b: float) -> float:
-    """
-    Multiplies two numbers a and b.
-    Use this tool for calculations involving multiplication.
-    """
-    return a * b
+    """Multiplies two numbers (a and b) and returns the result."""
+    return str(a * b)
 
-# FIX 2: Change task to 'text-generation'
-llama_llm = HuggingFaceInferenceAPI(
-    model_name="Qwen/Qwen2.5-7B-Instruct", 
-    token=HF_TOKEN,
-    task="text-generation" 
-)
+# Create the tools
+tools = [
+    FunctionTool.from_defaults(fn=get_tokyo_time),
+    FunctionTool.from_defaults(fn=multiply)
+]
 
+# Initialize Agent with a clear "React" header
 llama_agent = ReActAgent.from_tools(
-    [
-        FunctionTool.from_defaults(fn=get_tokyo_time),
-        FunctionTool.from_defaults(fn=multiply)
-    ],
+    tools,
     llm=llama_llm,
-    verbose=True
+    verbose=True # This will print the "Thought/Action" in your terminal/logs
 )
 
-# ============ GRADIO UI ============
-
-def chat_smol(message, history):
-    try:
-        return str(smol_agent.run(message))
-    except Exception as e:
-        return f"SmolAgent Error: {str(e)}"
+# ============ GRADIO UI (STABLE) ============
 
 def chat_llama(message, history):
     try:
-        # LlamaIndex keeps its own history, so we just pass the message
+        # ReAct agents can sometimes take 2-3 "steps" to answer
         response = llama_agent.chat(message)
         return str(response)
     except Exception as e:
-        return f"LlamaIndex Error: {str(e)}"
+        # This will tell you EXACTLY why it failed in the Chat window
+        return f"⚠️ LlamaIndex Error: {str(e)}"
 
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 🤖 Multi-Framework AI Agent Space")