Update app.py

app.py

@@ -10,20 +10,16 @@ LORA_PATH = "./"
 tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
 tokenizer.pad_token = tokenizer.eos_token
 
-# Base model (disk offload enabled)
-base_model = AutoModelForCausalLM.from_pretrained(
+# Base model (CPU)
+model = AutoModelForCausalLM.from_pretrained(
     BASE_MODEL,
     torch_dtype=torch.float32,
-    device_map="auto",
-    offload_folder="offload",     
-    offload_state_dict=True,
+    device_map={"": "cpu"},
     low_cpu_mem_usage=True
 )
 
-base_model.config.use_cache = False
-
-# Load LoRA (SADECE 1 KEZ)
-model = PeftModel.from_pretrained(base_model, LORA_PATH)
+# Load LoRA
+model = PeftModel.from_pretrained(model, LORA_PATH)
 model.eval()
 
 
@@ -45,7 +41,7 @@ Answer:
         output = model.generate(
             **inputs,
             max_new_tokens=int(max_tokens),
-            do_sample=False,
+            do_sample=False,          # CPU için hızlı
             eos_token_id=tokenizer.eos_token_id
         )
 
@@ -53,6 +49,7 @@ Answer:
     return tokenizer.decode(generated, skip_special_tokens=True)
 
 
+# Gradio UI
 demo = gr.Interface(
     fn=chat,
     inputs=[