squall/flash-attention
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Bump to v2.2.0

Browse Source
This commit is contained in:
Tri Dao 2023-09-05 11:34:13 -07:00
parent 8e893f0950
commit 6d673cd961
2 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
flash_attn/__init__.py
View File

@ -1,4 +1,4 @@
__version__ = "2.1.2.post3"
__version__ = "2.2.0"
from flash_attn.flash_attn_interface import (
flash_attn_func,

4
training/Dockerfile
View File

@ -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==2.1.2.post3
RUN pip install flash-attn==2.2.0
# Install CUDA extensions for cross-entropy, fused dense, layer norm
RUN git clone https://github.com/HazyResearch/flash-attention \
&& cd flash-attention && git checkout v2.1.2.post3 \
&& cd flash-attention && git checkout v2.2.0 \
&& cd csrc/fused_softmax && pip install . && cd ../../ \
&& cd csrc/rotary && pip install . && cd ../../ \
&& cd csrc/xentropy && pip install . && cd ../../ \