diff --git a/README.md b/README.md
index 96df7ba..87b7bb2 100644
--- a/README.md
+++ b/README.md
@@ -38,12 +38,11 @@ pip install flash-attn --no-build-isolation
 <details>
 <summary>Have trouble installing flash attention?</summary>
 
-You can create a clean conda env and install the following CUDA dev tools:
+In a clean conda env, install the `cuda-toolkit 12.3.0` for compilation:
 ```bash
-conda install conda-forge::cudatoolkit-dev -y
 conda install nvidia/label/cuda-12.3.0::cuda-toolkit -y
 ```
-Then you can install pytorch and flash attention:
+Then install pytorch and flash attention:
 ```
 # Install a torch version that you like.
 pip install torch==2.1.0 --index-url https://download.pytorch.org/whl/cu121