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

Set single threaded compilation for CUDA 12.2 so CI doesn't OOM

Browse Source
This commit is contained in:
Tri Dao 2023-09-03 23:41:05 -07:00
parent 6a89b2f121
commit 4976650f74
3 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.github/workflows/publish.yml vendored
View File

@ -151,7 +151,7 @@ jobs:
export PATH=/usr/local/nvidia/bin:/usr/local/nvidia/lib64:$PATH
export LD_LIBRARY_PATH=/usr/local/nvidia/lib64:/usr/local/cuda/lib64:$LD_LIBRARY_PATH
# Currently for this setting the runner goes OOM if we pass --threads 4 to nvcc
if [[ ${MATRIX_CUDA_VERSION} =~ "12." && ${MATRIX_TORCH_VERSION} == "2.1" ]]; then
if [[ ( ${MATRIX_CUDA_VERSION} == "121" || ${MATRIX_CUDA_VERSION} == "122" ) && ${MATRIX_TORCH_VERSION} == "2.1" ]]; then
export FLASH_ATTENTION_FORCE_SINGLE_THREAD="TRUE"
fi
# Limit MAX_JOBS otherwise the github runner goes OOM

2
flash_attn/__init__.py
View File

@ -1,4 +1,4 @@
__version__ = "2.1.2.post2"
__version__ = "2.1.2.post3"
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.post2
RUN pip install flash-attn==2.1.2.post3
# 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.post2 \
&& cd flash-attention && git checkout v2.1.2.post3 \
&& cd csrc/fused_softmax && pip install . && cd ../../ \
&& cd csrc/rotary && pip install . && cd ../../ \
&& cd csrc/xentropy && pip install . && cd ../../ \