diff --git a/flash_attn/__init__.py b/flash_attn/__init__.py
index 686e8d0f7..19950f89e 100644
--- a/flash_attn/__init__.py
+++ b/flash_attn/__init__.py
@@ -1,4 +1,4 @@
-__version__ = "2.3.0"
+__version__ = "2.3.1"
 
 from flash_attn.flash_attn_interface import (
     flash_attn_func,
diff --git a/training/Dockerfile b/training/Dockerfile
index f3ee6422c..ee2a5e79b 100644
--- a/training/Dockerfile
+++ b/training/Dockerfile
@@ -85,11 +85,11 @@ RUN pip install transformers==4.25.1 datasets==2.8.0 pytorch-lightning==1.8.6 tr
 RUN pip install git+https://github.com/mlcommons/logging.git@2.1.0
 
 # Install FlashAttention
-RUN pip install flash-attn==2.3.0
+RUN pip install flash-attn==2.3.1
 
 # Install CUDA extensions for fused dense, layer norm
 RUN git clone https://github.com/HazyResearch/flash-attention \
-    && cd flash-attention && git checkout v2.3.0 \
+    && cd flash-attention && git checkout v2.3.1 \
     && cd csrc/layer_norm && pip install . && cd ../../ \
     && cd csrc/fused_dense_lib && pip install . && cd ../../ \
     && cd .. && rm -rf flash-attention