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

Bump to v2.1.2

Browse Source
This commit is contained in:
Tri Dao 2023-09-03 21:49:20 -07:00
parent 0c04943fa2
commit 1dc1b6c8f2
4 changed files with 11 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
.github/workflows/publish.yml vendored
View File

@ -63,11 +63,17 @@ jobs:
# Pytorch <= 2.0 only supports CUDA <= 11.8
- torch-version: '1.12.1'
cuda-version: '12.1.0'
- torch-version: '1.12.1'
cuda-version: '12.2.0'
- torch-version: '1.13.1'
cuda-version: '12.1.0'
- torch-version: '1.13.1'
cuda-version: '12.2.0'
- torch-version: '2.0.1'
cuda-version: '12.1.0'
# Pytorch >= 2.1 only supports CUDA 12.1
- torch-version: '2.0.1'
cuda-version: '12.2.0'
# Pytorch >= 2.1 only supports CUDA >= 12.1
- torch-version: '2.1.0.dev20230731'
cuda-version: '11.6.2'
- torch-version: '2.1.0.dev20230731'

2
csrc/flash_attn/src/flash_fwd_kernel.h
View File

@ -947,7 +947,7 @@ inline __device__ void compute_attn_1rowblock_splitkv(const Params &params, cons
for (int mi = 0; mi < size<0>(acc_o_rowcol); ++mi) {
float sum = scores_sum(mi);
float inv_sum = (sum == 0.f || sum != sum) ? 1.f : 1.f / sum;
lse(mi) = (sum == 0.f || sum != sum) ? INFINITY : scores_max(mi) * params.scale_softmax + __logf(sum);
lse(mi) = (sum == 0.f || sum != sum) ? -INFINITY : scores_max(mi) * params.scale_softmax + __logf(sum);
float scale = inv_sum;
#pragma unroll
for (int ni = 0; ni < size<1>(acc_o_rowcol); ++ni) { acc_o_rowcol(mi, ni) *= scale; }

2
flash_attn/__init__.py
View File

@ -1,4 +1,4 @@
__version__ = "2.1.1"
__version__ = "2.1.2"
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==2.1.1
RUN pip install flash-attn==2.1.2
# Install CUDA extensions for cross-entropy, fused dense, layer norm
RUN git clone https://github.com/HazyResearch/flash-attention \
&& cd flash-attention && git checkout v2.1.1 \
&& cd flash-attention && git checkout v2.1.2 \
&& cd csrc/fused_softmax && pip install . && cd ../../ \
&& cd csrc/rotary && pip install . && cd ../../ \
&& cd csrc/xentropy && pip install . && cd ../../ \