diff --git a/olive/passes/pytorch/lora.py b/olive/passes/pytorch/lora.py
index 0b73a5472..9966b0b18 100644
--- a/olive/passes/pytorch/lora.py
+++ b/olive/passes/pytorch/lora.py
@@ -551,13 +551,14 @@ def get_peft_model(model: "PreTrainedModel", config: ConfigBase, config_kwargs:
         """Get the PEFT model for LoHa fine-tuning."""
         from peft import LoHaConfig, LoHaModel
 
+        target_modules = config.target_modules or "all-linear"
         config = LoHaConfig(
             r=config.r,
             alpha=config.alpha,
             rank_dropout=config.rank_dropout,
             module_dropout=config.module_dropout,
             use_effective_conv2d=config.use_effective_conv2d,
-            target_modules=config.target_modules,
+            target_modules=target_modules,
             exclude_modules=config.exclude_modules,
             init_weights=config.init_weights,
             layers_to_transform=config.layers_to_transform,
@@ -611,6 +612,7 @@ def get_peft_model(model: "PreTrainedModel", config: ConfigBase, config_kwargs:
         """Get the PEFT model for LoKr fine-tuning."""
         from peft import LoKrConfig, LoKrModel
 
+        target_modules = config.target_modules or "all-linear"
         config = LoKrConfig(
             r=config.r,
             alpha=config.alpha,
@@ -620,7 +622,7 @@ def get_peft_model(model: "PreTrainedModel", config: ConfigBase, config_kwargs:
             decompose_factor=config.decompose_factor,
             rank_dropout_scale=config.rank_dropout_scale,
             use_effective_conv2d=config.use_effective_conv2d,
-            target_modules=config.target_modules,
+            target_modules=target_modules,
             exclude_modules=config.exclude_modules,
             init_weights=config.init_weights,
             layers_to_transform=config.layers_to_transform,
@@ -864,7 +866,7 @@ def get_quant_model(
 
         # enable lora fine-tuning with the loftq initialized adapter weights
         pytorch_model = self.enable_lora(
-            pytorch_model, new_model_handler.task, config, adapter_path=loftq_init_adapter_path
+            pytorch_model, config, new_model_handler.task, adapter_path=loftq_init_adapter_path
         )
 
         return new_model_handler, pytorch_model, bnb_quant_config, quantized_modules
diff --git a/test/unit_test/passes/pytorch/test_lora.py b/test/unit_test/passes/pytorch/test_lora.py
index 94b6a4228..b5564081c 100644
--- a/test/unit_test/passes/pytorch/test_lora.py
+++ b/test/unit_test/passes/pytorch/test_lora.py
@@ -77,7 +77,8 @@ def run_finetuning(pass_class, tmp_path, **pass_config_kwargs):
     not torch.cuda.is_available() or True,
     reason="lora finetuning requires GPU.",
 )
-def test_lora(tmp_path):
+@patch("transformers.Trainer.train")
+def test_lora(mock_tain, tmp_path):
     # execute
     # bfloat16 is not supported on all gpu
     out = run_finetuning(LoRA, tmp_path, torch_dtype="float32")
@@ -91,7 +92,8 @@ def test_lora(tmp_path):
     platform.system() == OS.WINDOWS or not torch.cuda.is_available() or True,
     reason="bitsandbytes requires Linux GPU.",
 )
-def test_qlora(tmp_path):
+@patch("transformers.Trainer.train")
+def test_qlora(mock_tain, tmp_path):
     # execute
     # bfloat16 is not supported on all gpu
     out = run_finetuning(QLoRA, tmp_path, torch_dtype="float32")
@@ -105,7 +107,8 @@ def test_qlora(tmp_path):
     platform.system() == OS.WINDOWS or not torch.cuda.is_available() or True,
     reason="bitsandbytes requires Linux GPU.",
 )
-def test_loftq(tmp_path):
+@patch("transformers.Trainer.train")
+def test_loftq(mock_tain, tmp_path):
     # execute
     # bfloat16 is not supported on all gpu
     out = run_finetuning(LoftQ, tmp_path, torch_dtype="float32")