yindo/flash-attention-prebuild-wheels-rocm
1
0
Fork 0
mirror of https://github.com/BillyOutlast/flash-attention-prebuild-wheels-rocm.git synced 2026-07-01 01:37:53 -04:00
Code Issues Packages Projects Releases 2 Wiki Activity

update

Browse Source
This commit is contained in:
Junya Morioka
2025-08-07 10:01:44 +09:00
parent 8c829f3b31
commit e95088acb5
1 changed files with 28 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/workflows/build.yml
+28 -10
View File
@@ -21,13 +21,13 @@ jobs:
| Flash-Attention | Python | PyTorch | CUDA |
| --- | --- | --- | --- |
| 2.8.1 | 3.10, 3.11, 3.12 | 2.4.1, 2.5.1, 2.6.0, 2.7.1 | 12.4.1, 12.8.1 |
| 2.6.3, 2.7.4, 2.8.2 | 3.10, 3.11, 3.12 | 2.5.1, 2.6.0, 2.7.1, 2.8.0 | 12.4.1, 12.8.1, 12.9.1 |
## Windows x86_64
| Flash-Attention | Python | PyTorch | CUDA |
| --- | --- | --- | --- |
| 2.8.1 | 3.10, 3.11, 3.12 | 2.4.1, 2.5.1, 2.6.0, 2.7.1 | 12.4.1, 12.8.1 |
| 2.6.3, 2.7.4, 2.8.2 | 3.10, 3.11, 3.12 | 2.5.1, 2.6.0, 2.7.1, 2.8.0 | 12.4.1, 12.8.1, 12.9.1 |
"
# #########################################################
@@ -43,11 +43,11 @@ jobs:
strategy:
fail-fast: false
matrix:
flash-attn-version: ["2.8.1"]
flash-attn-version: ["2.6.3", "2.7.4", "2.8.2"]
python-version: ["3.10", "3.11", "3.12"]
torch-version: ["2.4.1", "2.5.1", "2.6.0", "2.7.1"]
torch-version: ["2.5.1", "2.6.0", "2.7.1", "2.8.0"]
# https://developer.nvidia.com/cuda-toolkit-archive
cuda-version: ["12.4.1", "12.8.1"]
cuda-version: ["12.4.1", "12.8.1", "12.9.1"]
exclude:
# torch < 2.2 does not support Python 3.12
- python-version: "3.12"
@@ -69,6 +69,13 @@ jobs:
# torch 2.7.0 does not support CUDA 12.4
- torch-version: "2.7.0"
cuda-version: "12.4.1"
# torch < 2.8 does not support CUDA 12.9
- torch-version: "2.5.1"
cuda-version: "12.9.1"
- torch-version: "2.6.3"
cuda-version: "12.9.1"
- torch-version: "2.7.1"
cuda-version: "12.9.1"
steps:
- uses: actions/checkout@v4
@@ -305,13 +312,11 @@ jobs:
strategy:
fail-fast: false
matrix:
flash-attn-version: ["2.8.1"]
flash-attn-version: ["2.6.3", "2.7.4", "2.8.2"]
python-version: ["3.10", "3.11", "3.12"]
torch-version: ["2.4.1", "2.5.1", "2.6.0", "2.7.1"]
torch-version: ["2.5.1", "2.6.0", "2.7.1", "2.8.0"]
# https://developer.nvidia.com/cuda-toolkit-archive
# CUDA 11.8 cannot build flash-attn.
# https://github.com/Dao-AILab/flash-attention/issues/595
cuda-version: ["12.4.1", "12.8.1"]
cuda-version: ["12.4.1", "12.8.1", "12.9.1"]
exclude:
# torch < 2.2 does not support Python 3.12
- python-version: "3.12"
@@ -327,9 +332,22 @@ jobs:
cuda-version: "12.6.3"
- torch-version: "2.0.1"
cuda-version: "12.8.1"
# torch 2.6.0 does not support CUDA 12.1
- torch-version: "2.6.0"
cuda-version: "12.1.1"
# torch 2.7.0 does not support CUDA 12.4
- torch-version: "2.7.0"
cuda-version: "12.4.1"
# flash-attn 2.7.4 does not build with CUDA 12.8
- flash-attn-version: "2.7.4"
cuda-version: "12.8.1"
# torch < 2.8 does not support CUDA 12.9
- torch-version: "2.5.1"
cuda-version: "12.9.1"
- torch-version: "2.6.3"
cuda-version: "12.9.1"
- torch-version: "2.7.1"
cuda-version: "12.9.1"
steps:
- uses: actions/checkout@v4