Update app.py

app.py

@@ -65,34 +65,15 @@ def generate(req: UserInputRequest):
             {"role": "user", "content": req.user_input}
         ]
 
-        # === Önce chat template string'ini al
-        chat_template_raw = tokenizer.apply_chat_template(
+        chat_input = tokenizer.apply_chat_template(
             messages,
             add_generation_prompt=True,
-            return_tensors=None
-        )
-
-        # Güvenlik: None veya beklenmedik tip gelirse zorla string'e çevir
-        if chat_template_raw is None:
-            chat_template_str = ""
-        elif isinstance(chat_template_raw, str):
-            chat_template_str = chat_template_raw
-        else:
-            chat_template_str = str(chat_template_raw)
-
-        # === Sonra tokenizer() ile input_ids + attention_mask hazırla
-        tokenized_inputs = tokenizer(
-            chat_template_str,
-            return_tensors="pt",
-            padding=True
+            return_tensors="pt"
         ).to(model.device)
 
-        input_ids = tokenized_inputs['input_ids']
-        attention_mask = tokenized_inputs['attention_mask']
-
-        input_len = input_ids.shape[-1]
+        input_len = chat_input.shape[-1]
         total_ctx = model.config.max_position_embeddings if hasattr(model.config, 'max_position_embeddings') else 4096
-        max_new_tokens = max(1, total_ctx - input_len)
+        max_new_tokens = min(512, max(1, total_ctx - input_len))
 
         log(f"ℹ️ Input uzunluğu: {input_len}, max_new_tokens ayarlandı: {max_new_tokens}")
 
@@ -102,8 +83,7 @@ def generate(req: UserInputRequest):
         ]
 
         outputs = model.generate(
-            input_ids=input_ids,
-            attention_mask=attention_mask,
+            input_ids=chat_input,
             max_new_tokens=max_new_tokens,
             eos_token_id=terminators
         )