截止目前最新的Linux&Windows系统平台上pip安装的xformers版本与pytorch版本和CUDA版本关系对照表
xformers | pytorch | CUDA |
v0.0.29.post2 | torch==2.6.0 | cu118,cu124,cu126 |
0.0.29.post1,0.0.29,0.0.28.post3 | torch==2.5.1 | cu118,cu121,cu124 |
0.0.28.post2 | torch==2.5.0 | cu118,cu121,cu124 |
0.0.28.post1 | torch==2.4.1 | cu118,cu121,cu124 |
0.0.27.post1,0.0.27.post2 | torch==2.3.1 | cu118,cu121 |
0.0.27 | torch==2.3.0 | cu118,cu121 |
0.0.26 | torch==2.2.0 | cu118,cu121 |
0.0.25 | torch==2.2.0 | cu118,cu121 |
0.0.24 | torch==2.2.0 | cu118,cu121 |
0.0.23 | torch==2.1.1 | cu118,cu121 |
0.0.22.post7 | torch==2.1.0 | cu118,cu121 |
0.0.22 | torch==2.0.1 | |
0.0.21 | torch==2.0.1 | |
0.0.20 | torch==2.0.1 | |
0.0.19 | torch==2.0.0 | |
0.0.18 | torch==2.0.0 | |
0.0.17 | torch==1.13.1 |
xformers v0.0.28版本Linux&Windows平台安装pip命令参考:
# [linux only] cuda 11.8 version
pip3 install -U xformers --index-url https://download.pytorch.org/whl/cu118
# [linux only] cuda 12.1 version
pip3 install -U xformers --index-url https://download.pytorch.org/whl/cu121
# [linux & win] cuda 12.4 version
pip3 install -U xformers --index-url https://download.pytorch.org/whl/cu124
# [linux only] (EXPERIMENTAL) rocm 6.1 version
pip3 install -U xformers --index-url https://download.pytorch.org/whl/rocm6.1
安装xformers的时候可以和torch一起安装,不然的话torch还会再重新安装一次,而且torch会自动选择到当前xformers支持到的最新版本,可能会与你当前想要的torch版本不一样导致torchvision等版本冲突,最好指定版本安装。例如指定安装最新版torch 2.5.1和xformers 0.0.28.post3和cu124,可以使用如下安装命令:
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 xformers==0.0.28.post3 --index-url https://download.pytorch.org/whl/cu124
附:
xformers whl下载链接:https://download.pytorch.org/whl/xformers/
xformers项目地址:https://github.com/facebookresearch/xformers
目前针对torch 2.7.0+CUDA128官方还没有发布xformers正式编译版本,我以前也没自己构建过xformers,今天试了下,我I7 CPU+64G内存的电脑都瞬间跑满而且卡掉了,太难了。
你想用的话可以先用着下面这个xformers 0.0.30版本,此版本只支持python 3.12
xformers 0.0.30版本下载链接:https://pan.quark.cn/s/67138900fef2