Update llm_model.py

llm_model.py

@@ -12,30 +12,30 @@ eos_token_id = None
 class Message(BaseModel):
     user_input: str
 
-def setup_model(app_config):
+def setup_model(service_config):
     try:
         log("🧠 setup_model() başladı")
         device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         log(f"📡 Kullanılan cihaz: {device}")
-        tokenizer = AutoTokenizer.from_pretrained(app_config.MODEL_BASE, use_fast=False)
+        tokenizer = AutoTokenizer.from_pretrained(service_config.MODEL_BASE, use_fast=False)
         log("📦 Tokenizer yüklendi. Ana model indiriliyor...")
-        model = AutoModelForCausalLM.from_pretrained(app_config.MODEL_BASE, torch_dtype=torch.float32).to(device)
+        model = AutoModelForCausalLM.from_pretrained(service_config.MODEL_BASE, torch_dtype=torch.float32).to(device)
         log("📦 Ana model indirildi ve yüklendi. eval() çağırılıyor...")
         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()
         log("✅ Ana model eval() çağrıldı")
-        log(f"📦 Intent modeli indiriliyor: {app_config.INTENT_MODEL_ID}")
-        _ = AutoTokenizer.from_pretrained(app_config.INTENT_MODEL_ID)
-        _ = AutoModelForSequenceClassification.from_pretrained(app_config.INTENT_MODEL_ID)
+        log(f"📦 Intent modeli indiriliyor: {service_config.INTENT_MODEL_ID}")
+        _ = AutoTokenizer.from_pretrained(service_config.INTENT_MODEL_ID)
+        _ = AutoModelForSequenceClassification.from_pretrained(service_config.INTENT_MODEL_ID)
         log("✅ Intent modeli önbelleğe alındı.")
         log("✔️ Model başarıyla yüklendi ve sohbet için hazır.")
     except Exception as e:
         log(f"❌ setup_model() hatası: {e}")
         traceback.print_exc()
 
-async def generate_response(text, app_config):
+async def generate_response(text, service_config):
     messages = [{"role": "user", "content": text}]
     encodeds = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True)
     eos_token = tokenizer("<|im_end|>", add_special_tokens=False)["input_ids"][0]
@@ -47,14 +47,14 @@ async def generate_response(text, app_config):
             input_ids=input_ids,
             attention_mask=attention_mask,
             max_new_tokens=128,
-            do_sample=app_config.USE_SAMPLING,
+            do_sample=service_config.USE_SAMPLING,
             eos_token_id=eos_token,
             pad_token_id=tokenizer.pad_token_id,
             return_dict_in_generate=True,
             output_scores=True
         )
 
-    if not app_config.USE_SAMPLING:
+    if not service_config.USE_SAMPLING:
         scores = torch.stack(output.scores, dim=1)
         probs = torch.nn.functional.softmax(scores[0], dim=-1)
         top_conf = probs.max().item()