Update inference_test_turkcell_with_intents.py

inference_test_turkcell_with_intents.py

@@ -165,7 +165,11 @@ async def generate_response(text):
             output_scores=True
         )
 
-    return tokenizer.decode(output.sequences[0], skip_special_tokens=True).strip()
+    try:
+        return tokenizer.decode(output.sequences[0], skip_special_tokens=True).strip()
+    except Exception as decode_error:
+        log(f"❌ Decode hatası: {decode_error}")
+        return random.choice(FALLBACK_ANSWERS)
 
 @app.post("/chat")
 async def chat(msg: Message):
@@ -192,17 +196,25 @@ async def chat(msg: Message):
         traceback.print_exc()
         return JSONResponse(content={"error": str(e)}, status_code=500)
 
+def log(message):
+    timestamp = datetime.now().strftime("%H:%M:%S")
+    print(f"[{timestamp}] {message}", flush=True)
+
 def setup_model():
     global model, tokenizer, eos_token_id
     try:
+        log("🧠 setup_model() başladı")
         device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        log("📥 Tokenizer indiriliyor...")
         tokenizer = AutoTokenizer.from_pretrained(MODEL_BASE, use_fast=False)
+        log("🧠 Model indiriliyor...")
         model = AutoModelForCausalLM.from_pretrained(MODEL_BASE, torch_dtype=torch.float32).to(device)
         tokenizer.pad_token = tokenizer.pad_token or tokenizer.eos_token
         model.config.pad_token_id = tokenizer.pad_token_id
         eos_token_id = tokenizer("<|im_end|>", add_special_tokens=False)["input_ids"][0]
         model.eval()
-        print("[✔️] Model başarıyla yüklendi ve sohbet için hazır.")
+        log("✅ Model eval() çağrıldı")
+        log("✔️ Model başarıyla yüklendi ve sohbet için hazır.")
     except Exception as e:
         traceback.print_exc()