Update app.py

app.py

@@ -4,6 +4,12 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 import torch
 import os
 import threading
+import time
+import logging
+
+#Cấu hình logging
+logging.basicConfig(level=logging.INFO)
+
 
 app = FastAPI()
 
@@ -19,6 +25,7 @@ model = AutoModelForCausalLM.from_pretrained(
     device_map="auto",
     use_auth_token=HF_TOKEN
 )
+logging.info("Mô hình đã sẵn sàng!")
 @app.get("/")
 def read_root():
     return {"message": "Ứng dụng đang chạy!"}
@@ -42,7 +49,7 @@ async def chat(request: ChatRequest):
     # Sinh phản hồi từ mô hình
     outputs = model.generate(
         input_ids=input_ids,
-        max_new_tokens=100,
+        max_new_tokens=80,
         temperature=0.7,
         top_k=50,
         top_p=0.9,
@@ -66,13 +73,24 @@ def keep_model_alive():
                 add_generation_prompt=True,
                 return_tensors="pt"
             ).to(model.device)
+            with torch.inference_mode():
+                _ = model.generate(inputs, max_new_tokens=5)
 
-            _ = model.generate(**inputs, max_new_tokens=10)
+            logging.info("Heartbeat OK")
         except Exception as e:
-            print("Heartbeat error:", e)
+            logging.error(f"Heartbeat error: {e}")
         time.sleep(300)  # Gọi mỗi 5 phút
 
+
+
+
 # Khởi động tác vụ nền khi app chạy
+
 @app.on_event("startup")
-def startup_event():
-    threading.Thread(target=keep_model_alive, daemon=True).start()
+async def startup_event():
+    # Dùng background task thay vì thread
+    from asyncio import create_task
+    create_task(run_keep_alive())
+
+async def run_keep_alive():
+    keep_model_alive()