Posted on | In

Source

Good reference 安裝 mamba: https://zhuanlan.zhihu.com/p/25916604332

用 nvcc check cuda 版本，不要用 nvidia-smi! 我用 cuda 12.1, python 3.10, pytorch 2.5!

conda create –name mamba2 python=3.10 conda activate mamba2 2246 lrt 2247 pip install torch torchvision torchaudio –index-url https://download.pytorch.org/whl/cu121

Problem 1: Flash attention!

pre-built, conda 都不行。最後這版好像 ok.

1
pip install flash-attn==2.5.8 --no-build-isolation

安裝: causal_conv1d!

使用命令查看是否支持abi python -c “import torch; print(torch._C._GLIBCXX_USE_CXX11_ABI)” 我的输出 false，就选择abiFALSE的 记得点show all选择对应的torch版本 pip install tools/causal_conv1d-1.5.2+cu12torch2.5cxx11abiFALSE-cp310-cp310-linux_x86_64.whl

做完後 run pytest: 8794 passed, 3888 skipped in 133.40s

From llama3 開始 source code installation

https://blog.csdn.net/yyywxk/article/details/144790950

git clone https://github.com/Dao-AILab/causal-conv1d.git cd causal-conv1d git checkout v1.1.1 # 用你想要的版本 CAUSAL_CONV1D_FORCE_BUILD=TRUE pip install . cd .. git clone https://github.com/state-spaces/mamba.git cd mamba git checkout v1.1.1 # 用你想要的版本 pip install . # 方式一，下载whl安装，两种方式选择一个即可 MAMBA_FORCE_BUILD=TRUE pip install . # 方式二，强制在本地编译安装，Win 下无法识别此命令

Problem 3: upgrade mamba-ssm pip install –upgrade mamba-ssm? (from 1.1 to 2.2)