Update app.py

app.py

@@ -28,8 +28,10 @@ model, tokenizer = FastLanguageModel.from_pretrained(
     max_seq_length=max_seq_length,
     dtype=dtype,
     load_in_4bit=load_in_4bit,
-    token=hf_token
+    token=hf_token,
+    load_in_8bit_fp32_cpu_offload=True,  # Enable CPU offloading
 )
+
 print("Model and tokenizer loaded successfully.")
 
 print("Configuring PEFT model...")
@@ -119,7 +121,7 @@ trainer = SFTTrainer(
         bf16=is_bfloat16_supported(),
         warmup_steps=5,
         logging_steps=10,
-        max_steps=100,
+        max_steps=50,
         optim="adamw_8bit",
         weight_decay=0.01,
         lr_scheduler_type="linear",