perf: Preload model on container startup and optimize PyTorch CPU multi-threading

app.py

@@ -7,6 +7,7 @@ import os
 import shutil
 import sys
 import time
+from contextlib import asynccontextmanager
 from pathlib import Path
 from typing import Any
 
@@ -25,7 +26,15 @@ from controlai_rag.chunker import chunk_document
 from controlai_rag.document_loader import load_single_file
 from controlai_rag.index import ControlRAGIndex
 
-app = FastAPI(title="ControlAI", version="1.0.0")
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    print("Pre-loading ControlAI Core Engine on startup...")
+    get_agent()
+    print("ControlAI Core Engine is online and ready for traffic.")
+    yield
+
+
+app = FastAPI(title="ControlAI", version="1.0.0", lifespan=lifespan)
 
 app.add_middleware(
     CORSMiddleware,

controlai_agent/orchestrator.py

@@ -183,11 +183,23 @@ class ControlAIAgent:
             self.hf_tokenizer = AutoTokenizer.from_pretrained(model_path)
         else:
             # Universal PyTorch / Transformers fallback on Linux, Colab, HuggingFace, CUDA
+            import torch
+            num_threads = min(os.cpu_count() or 2, 4)
+            try:
+                torch.set_num_threads(num_threads)
+            except Exception:
+                pass
+
             hf_id = "Qwen/Qwen2.5-3B-Instruct" if "mlx" in str(model_path) else model_path
+            print(f"Loading PyTorch model: {hf_id} (CPU threads: {num_threads})...")
             self.hf_tokenizer = AutoTokenizer.from_pretrained(hf_id, trust_remote_code=True)
+            
+            # Use bfloat16 on CUDA, float32 on CPU for clean numerical stability
+            dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32
             self.model = AutoModelForCausalLM.from_pretrained(
                 hf_id,
-                torch_dtype="auto",
+                torch_dtype=dtype,
+                low_cpu_mem_usage=True,
                 device_map="auto",
                 trust_remote_code=True,
             )
@@ -195,8 +207,10 @@ class ControlAIAgent:
                 try:
                     from peft import PeftModel
                     self.model = PeftModel.from_pretrained(self.model, adapter_path)
-                except Exception:
-                    pass
+                    print(f"Loaded PEFT LoRA adapter from: {adapter_path}")
+                except Exception as exc:
+                    print(f"Warning: Could not load LoRA adapter in PyTorch: {exc}")
+            print("PyTorch model loaded successfully.")
 
         # Initialize local offline RAG index
         try: