diff --git a/docs/source/conf.py b/docs/source/conf.py
index 8408041..e2f01a3 100644
--- a/docs/source/conf.py
+++ b/docs/source/conf.py
@@ -52,6 +52,7 @@ extensions = [
     'sphinx.ext.autosummary',
     'sphinx.ext.doctest',
     'sphinx.ext.intersphinx',
+    # 'sphinx.ext.mathbase',
     'sphinx.ext.mathjax',
     'sphinx.ext.napoleon',
     'sphinx.ext.viewcode',
diff --git a/examples/examples_text-classification/configs/lora_roberta-base/lora_mrpc.json b/examples/examples_text-classification/configs/lora_roberta-base/lora_mrpc.json
index 46afef5..10eeb38 100644
--- a/examples/examples_text-classification/configs/lora_roberta-base/lora_mrpc.json
+++ b/examples/examples_text-classification/configs/lora_roberta-base/lora_mrpc.json
@@ -38,7 +38,8 @@
     "tokenizer_name": "roberta-base",
     "unfrozen_modules": [
         "classifier",
-        "deltas"
+        "deltas",
+        "layer_norm"
     ],
     "warmup_ratio": 0.06,
     "weight_decay": 0.1,
diff --git a/opendelta/delta_models/lora.py b/opendelta/delta_models/lora.py
index 492fea6..02ebe85 100644
--- a/opendelta/delta_models/lora.py
+++ b/opendelta/delta_models/lora.py
@@ -66,8 +66,13 @@ class LoraConfig(BaseDeltaConfig):
 
 class LoraModel(DeltaBase):
     r""" The implementation of `LoRA: Low-Rank Adaptation of Large Language Models <https://arxiv.org/abs/2106.09685>`_ .
-    Thanks for their `loralib <https://github.com/microsoft/LoRA/tree/main/loralib>`_, we use loralib.linear 
-    to replace the linear layer of the backbone model. 
+    Thanks for their `loralib <https://github.com/microsoft/LoRA/tree/main/loralib>`_.
+    
+    .. note::
+        In our implementation, we did not use loralib.linear to replace the linear layer of the backbone model.
+        Instead, we insert a parallel module into the backbone.
+        In other words, we treat :math:`(W + A^TB) X` as :math:`WX+ A^TBX`, and insert the :math:`A^TBX` as a parallel insertion module. 
+        If you want to use the original implementation, please refer to `lora_old.py`
 
     class attributes:
         - default_modified_modules = ['attn.q', 'attn.v'] According to the paper, they modify q and v matrix in the