I also come to share my environment and methods

[shabbyliu]

my version

cuda = 11.8
python = 3.10.13
gcc = (Ubuntu 11.4.0-9ubuntu1) 11.4.0
g++ = (Ubuntu 11.4.0-9ubuntu1) 11.4.0
Ninja = 1.11.1.git.kitware.jobserver-1
Ubuntu = 24.04

Make sure gcc, g++, and Ninja are installed

$ sudo apt update
$ sudo apt install build-essential
$ sudo apt install ninja-build
$ gcc --version
$ g++ --version
$ ninja --version

cmd input

$ conda create -n vim python=3.10.13

$ conda activate vim

$ pip install torch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 --index-url https://download.pytorch.org/whl/cu118

$ cd ....../Vim-main/

$ pip install -r vim/vim_requirements.txt

$ pip install causal-conv1d==1.1.0

$ cd ....../Vim-main/mamba-1p1p1

$ pip install -e .

test in python

from mamba_ssm import Mamba
batch, length, dim = 2, 64, 16
x = torch.randn(batch, length, dim).to("cuda")
model = Mamba(
# This module uses roughly 3 * expand * d_model^2 parameters
d_model=dim, # Model dimension d_model
d_state=16, # SSM state expansion factor
d_conv=4, # Local convolution width
expand=2, # Block expansion factor
).to("cuda")
y = model(x)
assert y.shape == x.shape

Modify function in selective_scan_interface.py on line 324

conv1d_out = causal_conv1d_cuda.causal_conv1d_fwd(
x, conv1d_weight, conv1d_bias, None, None, None, True
)
to
conv1d_out = causal_conv1d_cuda.causal_conv1d_fwd(
x, conv1d_weight, conv1d_bias, None, True
)