安装Mamba问题

问题1ImportError causal_conv1d_cuda.cpython-38-x86_64-linux-gnu.so undefined symbol

这个有直接在https://github.com/Dao-AILab/causal-conv1d/releases/tag/v1.0.1 尝试下载whl

再在下载目录安装对应的

pip install  causal_conv1d-1.0.1+cu118torch2.2cxx11abiFALSE-cp38-cp38-linux_x86_64.whl 

之后在https://github.com/state-spaces/mamba/releases/tag/v1.0.1

用同样的方式

pip install  mamba_ssm-1.0.1+cu118torch2.2cxx11abiFALSE-cp38-cp38-linux_x86_64.whl

但是报错如上，因此参考博客安装mamba错误ImportError causal_conv1d_cuda.cpython-310-x86_64-linux-gnu.so undefined symbol_windows系统怎么安装causal-conv1d-CSDN博客

git clone https://github.com/Dao-AILab/causal-conv1d.gitz

再直接分布安装casual-conv1d和mamba-ssm

pip install .
pip install mamba-ssm

成功解决该问题，但会产生以下报错

问题2

TypeError: causal_conv1d_fwd(): incompatible function arguments. The following argument types are supported:     1. (arg0: torch.Tensor, arg1: torch.Tensor, arg2: Optional[torch.Tensor], arg3: bool) -> torch.Tensor

发现是因为causal-conv1d版本不够，因此在通过git clone https://github.com/Dao-AILab/causal-conv1d.gitz安装的causal-conv1d目录下

git checkout v1.1.1

再安装casual-conv1d和mamba-ssm

pip install .
pip install mamba-ssm

通过

from mamba_ssm import Mamba 

a = Mamba(d_model=256)

可以成功编译运行