Vim-Light

通用环境配置

git clone https://github.com/HEUfcy/Vim-Light.git
cd Vim-Light

conda create -n visionmamba python=3.10
conda activate visionmamba
conda install -c nvidia cuda-nvcc=11.8.89 # nvcc版本在11.8，CUDA版本应高于11.8
pip install torch==2.1.1+cu118 torchvision==0.16.1+cu118 torchaudio==2.1.1+cu118 --index-url https://download.pytorch.org/whl/cu118 # 版本很重要
conda install cudatoolkit=11.8 # 适合pytorch: 2.0.0，2.0.1，2.1.0，2.1.1，2.1.2，2.2.0，2.2.1
pip install xformers==0.0.23
pip install numpy==1.26 -i https://pypi.tuna.tsinghua.edu.cn/simple

xformers对应torch版本 （选配）

xformers	 pytorch
0.0.26.post1  torch==2.3.0
0.0.25	       torch==2.2.1
0.0.24	       torch==2.2.0
0.0.23	       torch==2.1.1
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==2.0.0
0.0.16	       torch==1.13.1

安装deepspeed （选配）

pip install deepspeed
# 当安装报错：FileNotFoundError: [Errno 2] No such file or directory: '/usr/local/cuda-11.7/bin/nvcc'时
先运行：export CUDA_HOME=/usr/local/cuda
# 参考：https://blog.csdn.net/BetrayFree/article/details/132598503

Vision Mamba环境配置

安装包下载到本地再上传

wget https://github.com/Dao-AILab/causal-conv1d/releases/download/v1.1.3.post1/causal_conv1d-1.1.3.post1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
wget https://github.com/state-spaces/mamba/releases/download/v1.1.1/mamba_ssm-1.1.1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
pip install causal_conv1d-1.1.3.post1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl 
pip install mamba_ssm-1.1.1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

先安装causal_sonvld-1.1.3

pip install causal_conv1d-1.1.3.post1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
# 检查是否安装成功
import torch
import causal_conv1d_cuda

再安装mamba_ssm-1.1.1

pip install mamba_ssm-1.1.1+cu118torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl
# 检查是否安装成功
import torch
import mamba_ssm

检查环境

python Vim.py
# 运行得到：
# preds shape if torch.Size([4, 1000])