diff --git a/lerobot/common/policies/dexvla/README.md b/lerobot/common/policies/dexvla/README.md
index 9d0b9805..b34a40bb 100644
--- a/lerobot/common/policies/dexvla/README.md
+++ b/lerobot/common/policies/dexvla/README.md
@@ -2,7 +2,7 @@
 DexVLA: Vision-Language Model with Plug-In Diffusion Expert for Visuomotor Policy Learning</h1>
 
 This policy is Community Contributed. For more information about DexVLA, you can also refer to [this](https://github.com/juruobenruo/DexVLA).
-This is [project website](https://dex-vla.github.io/). 
+This is [project website](https://dex-vla.github.io/).
 
 ## Dataset
 ### Data format
@@ -141,4 +141,4 @@ python lerobot/scripts/eval.py \
 ~~~
 
 ### Inference Speed
-Tested on a single A6000 GPU, the DexVLA could infer 3.4 action chunks in one second. For each action chunk, if we execute 25 actions, the real control frequency can be 85 (3.4*25)Hz.
\ No newline at end of file
+Tested on a single A6000 GPU, the DexVLA could infer 3.4 action chunks in one second. For each action chunk, if we execute 25 actions, the real control frequency can be 85 (3.4*25)Hz.
diff --git a/lerobot/common/policies/dexvla/fusion_modules.py b/lerobot/common/policies/dexvla/fusion_modules.py
index bff61072..701a4ada 100644
--- a/lerobot/common/policies/dexvla/fusion_modules.py
+++ b/lerobot/common/policies/dexvla/fusion_modules.py
@@ -16,6 +16,7 @@
 
 import torch.nn as nn
 
+
 class ActionProjector(nn.Module):
     def __init__(self, in_dim, out_dim=1024):
         super().__init__()
diff --git a/lerobot/common/policies/dexvla/robot_data_processor.py b/lerobot/common/policies/dexvla/robot_data_processor.py
index 3a7ce995..eba11890 100644
--- a/lerobot/common/policies/dexvla/robot_data_processor.py
+++ b/lerobot/common/policies/dexvla/robot_data_processor.py
@@ -19,6 +19,7 @@ import torch
 from PIL import Image
 from qwen_vl_utils import fetch_image
 
+
 class Qwen2VLAProcess:
     def __init__(
         self,