Update app.py

app.py

@@ -27,16 +27,20 @@ if not os.path.exists(global_context_path):
     with open(global_context_path, "w") as f:
         f.write("# Global Context & Browser Workspace\nShared knowledge base accessible across all connected IP environments under /px.")
 
-# Zero-dependency auto-bootstrapper
-for package in ["gradio", "requests", "ollama", "beautifulsoup4"]:
+# Auto-bootstrapper for ZeroGPU environment requirements
+required_packages = ["gradio==4.44.1", "requests", "spaces", "torch", "llama-cpp-python", "beautifulsoup4"]
+for package in required_packages:
     try:
-        __import__(package)
+        pkg_name = package.split("==")[0]
+        __import__(pkg_name)
     except ImportError:
         print(f"Installing missing dependency: {package}...")
         subprocess.check_call([sys.executable, "-m", "pip", "install", "-q", package])
 
 import requests
-import ollama
+import spaces
+import torch
+from llama_cpp import Llama
 import gradio as gr
 from bs4 import BeautifulSoup
 
@@ -65,14 +69,30 @@ with open(LOG_FILE, "a") as log_file:
     log_file.write(log_entry)
 
 print("\n" + "=" * 65)
-print(f"🌐 PX TECH SOLUTIONS - EVAL BETA 0.2 AGENTIC CORE (/px)")
+print(f"🌐 PX TECH SOLUTIONS - EVAL BETA 0.2 ZEROGPU CORE (/px)")
 print(f"   • Base Directory : {BASE_DIR}")
 print(f"   • Local URL      : http://{local_ip}:7860")
 print(f"   • Server IP      : {server_public_ip}")
-print(f"   • Active Engine  : Bash | Write | Read | Search | Big Context")
+print(f"   • Active Engine  : ZeroGPU (@spaces.GPU) | Bash | Write | Read | Search")
 print("=" * 65 + "\n")
 
-MODEL_NAME = "smarthomemain10123/eval-beta-0.2"
+# Model Loader (Lazy initialized inside ZeroGPU context)
+MODEL_PATH = os.path.join(BASE_DIR, "eval-beta-0.2.gguf")
+_llm_instance = None
+
+def get_llm():
+    global _llm_instance
+    if _llm_instance is None:
+        print(f"📦 Initializing model weights from {MODEL_PATH} on ZeroGPU...")
+        if not os.path.exists(MODEL_PATH):
+            print("⚠️ Warning: Model GGUF file not found in /px. Please ensure your model weights are uploaded.")
+        _llm_instance = Llama(
+            model_path=MODEL_PATH if os.path.exists(MODEL_PATH) else "model.gguf",
+            n_ctx=4096,
+            n_gpu_layers=-1, # Full GPU offload via ZeroGPU
+            verbose=False
+        )
+    return _llm_instance
 
 # Custom Cyber-Dark UI Styling
 custom_css = """
@@ -81,7 +101,7 @@ body { background-color: #0b0f19; color: #f3f4f6; }
 """
 
 # =====================================================================
-# AGENTIC TOOL DEFINITIONS (OpenCode Ecosystem)
+# AGENTIC TOOL DEFINITIONS
 # =====================================================================
 tools_definition = [
     {
@@ -137,7 +157,6 @@ tools_definition = [
     }
 ]
 
-# Tool Executors
 def execute_tool(name, args, ip_folder, yolo_mode):
     try:
         if name == "bash":
@@ -181,8 +200,9 @@ def execute_tool(name, args, ip_folder, yolo_mode):
         return f"Tool execution failed: {str(e)}"
 
 # =====================================================================
-# AGENT PREDICT LOOP
+# ZEROGPU DECORATED PREDICT LOOP
 # =====================================================================
+@spaces.GPU(duration=120)
 def predict(message, history, system_prompt, temperature, yolo_mode, request: gr.Request):
     client_ip = "127.0.0.1"
     if request and request.headers:
@@ -221,63 +241,31 @@ def predict(message, history, system_prompt, temperature, yolo_mode, request: gr
         f"--- [USER IP CONTEXT] ---\nClient IP: {client_ip} | Workspace Folder: {ip_folder} | Security Mode: {security_mode}"
     )
 
-    messages = [{"role": "system", "content": enhanced_system_prompt}]
+    prompt = f"System: {enhanced_system_prompt}\n"
     for human, assistant in history:
-        messages.append({"role": "user", "content": human})
-        messages.append({"role": "assistant", "content": assistant})
-    messages.append({"role": "user", "content": message})
+        prompt += f"<|im_start|>user\n{human}<|im_end|>\n<|im_start|>assistant\n{assistant}<|im_end|>\n"
+    prompt += f"<|im_start|>user\n{message}<|im_end|>\n<|im_start|>assistant\n"
 
     history_log_path = os.path.join(ip_folder, "chat_history.log")
     with open(history_log_path, "a") as hf:
         hf.write(f"[{timestamp}] User (YOLO: {yolo_mode} | Sec: {security_mode}): {message}\n")
 
-    max_turns = 5
-    turn = 0
-    accumulated_output = ""
+    llm = get_llm()
+    
+    # Stream generation
+    output = llm(
+        prompt,
+        max_tokens=1024,
+        stop=["<|im_end|>", "<|im_start|>", "User:"],
+        temperature=temperature,
+        stream=True
+    )
 
-    while turn < max_turns:
-        turn += 1
-        try:
-            response = ollama.chat(
-                model=MODEL_NAME,
-                messages=messages,
-                tools=tools_definition,
-                options={"temperature": temperature}
-            )
-            
-            message_resp = response.get("message", {})
-            content = message_resp.get("content", "")
-            tool_calls = message_resp.get("tool_calls", [])
-
-            if content:
-                accumulated_output += content
-                yield accumulated_output
-
-            if tool_calls:
-                messages.append(message_resp)
-                for tc in tool_calls:
-                    func = tc.get("function", {})
-                    func_name = func.get("name")
-                    func_args = func.get("arguments", {})
-                    
-                    status_msg = f"\n\n⚙️ `[Executing Tool: {func_name} | Args: {json.dumps(func_args)}]`...\n"
-                    accumulated_output += status_msg
-                    yield accumulated_output
-
-                    tool_result = execute_tool(func_name, func_args, ip_folder, yolo_mode)
-                    
-                    messages.append({
-                        'role': 'tool',
-                        'content': tool_result
-                    })
-                    
-                    accumulated_output += f"```output\n{tool_result}\n```\n"
-                    yield accumulated_output
-            else:
-                break
-        except Exception as e:
-            yield accumulated_output + f"\n\n⚠️ **Agent Execution Error:** {str(e)}"
-            break
+    accumulated_output = ""
+    for chunk in output:
+        delta = chunk["choices"][0]["text"]
+        accumulated_output += delta
+        yield accumulated_output
 
     with open(history_log_path, "a") as hf:
         hf.write(f"[{timestamp}] Assistant: {accumulated_output}\n")
@@ -293,18 +281,18 @@ def update_global_context(new_content):
 with gr.Blocks() as demo:
     gr.Markdown(
         """
-        # ⚡ EVAL BETA 0.2 | PX TECH SOLUTIONS (AGENTIC OPENCODE ENVIRONMENT)
-        ### Autonomous Environment with Full Tool-Use (Bash, File Read/Write, Web Search, Big Context) & YOLO Security Mode
+        # ⚡ EVAL BETA 0.2 | PX TECH SOLUTIONS (ZEROGPU AGENTIC ENVIRONMENT)
+        ### Autonomous Environment with ZeroGPU Acceleration, Full Tool-Use, & YOLO Security Mode
         """
     )
     
     with gr.Row():
-        gr.Markdown(f"🟢 **Status:** Active | 📁 **Root Path:** `{BASE_DIR}` | ⏱️ **Slot 1 Active (48h Limit)** | 🛠️ **Tools Enabled:** Bash, Write, Read, Search")
+        gr.Markdown(f"🟢 **Status:** Active (ZeroGPU) | 📁 **Root Path:** `{BASE_DIR}` | ⏱️ **Slot 1 Active (48h Limit)** | 🛠️ **Tools Enabled:** Bash, Write, Read, Search")
 
     with gr.Accordion("⚙️ Engine Configuration, Global Context & YOLO Security Toggle", open=False):
         system_prompt_input = gr.Textbox(
             label="System Environment Prompt",
-            value="You are EVAL BETA 0.2, an advanced autonomous agent with full tool capabilities (bash execution, file manipulation, big context inspection, web searching) created by Pripro / PX TECH SOLUTIONS.",
+            value="You are EVAL BETA 0.2, an advanced autonomous agent with full tool capabilities created by Pripro / PX TECH SOLUTIONS.",
             lines=2
         )
         temperature_slider = gr.Slider(
@@ -334,14 +322,13 @@ with gr.Blocks() as demo:
     gr.Markdown(
         """
         ---
-        **Queue Architecture & /px Layout:** Strict queue line-up (`concurrency_limit=1`) with Slot 1 allocation and 48-hour session limits. All tools operate within isolated workspace paths under `/px`.
+        **Queue Architecture & /px Layout:** Strict queue line-up (`concurrency_limit=1`) with Slot 1 allocation and 48-hour session limits. Powered by Hugging Face ZeroGPU (`@spaces.GPU`).
         """
     )
 
 if __name__ == "__main__":
-    print(f"🚀 Launching agentic environment under {BASE_DIR} with Slot 1 Queue & Full Tools on port 7860...")
+    print(f"🚀 Launching ZeroGPU environment under {BASE_DIR} with Slot 1 Queue on port 7860...")
     demo.queue(default_concurrency_limit=1, max_size=20)
-    # Gradio 6 compliant theme and css passed directly to launch()
     demo.launch(
         server_name="0.0.0.0", 
         server_port=7860,