Update app.py

app.py

@@ -29,6 +29,21 @@ MODELS = [
     'Aravindan/awesome-gpt-2-coder', 'Qwen/Qwen2.5-Coder-0.5B'
 ]
 
+ACTIVE_SESSIONS = {}
+SESSION_TIMEOUT = 5*60
+
+def live_count(request: gr.Request):
+    current_time = time.time()
+    if request:
+        ACTIVE_SESSIONS[request.session_hash] = current_time
+        
+    # Prune
+    expired = [s for s, t in ACTIVE_SESSIONS.items() if current_time - t > SESSION_TIMEOUT]
+    for s in expired:
+        ACTIVE_SESSIONS.pop(s, None)
+        
+    return len(ACTIVE_SESSIONS)
+
 # Global class to safely manage the loaded model and tokenizer in memory
 class ModelManager:
     def __init__(self):
@@ -42,10 +57,11 @@ def get_system_stats():
     """Returns a dictionary of current system metrics with formatted strings."""
     mem = psutil.virtual_memory()
     disk = psutil.disk_usage('/')
-    return '\n'.join((
-        f"CPU\t: {psutil.cpu_percent(interval=1)}%",
-        f"Memory\t: {round(mem.used / (1024**3), 2)} / {round(mem.total / (1024**3), 2)} GB",
-        f"Disk\t: {round(disk.used / (1024**3), 2)} / {round(disk.total / (1024**3), 2)} GB"
+    return (
+        f"CPU\t\t: {psutil.cpu_percent(interval=1)}%\n"
+        f"Mem\t\t: {round(mem.used / (1024**3), 2)} / {round(mem.total / (1024**3), 2)} GB\n"
+        f"Disk\t\t: {round(disk.used / (1024**3), 2)} / {round(disk.total / (1024**3), 2)} GB\n"
+        f"Active\t\t: {live_count()}/5min"
     ))
 
 def load_new_model(model_id):