Unable to compile mamba under Windows ?

[juntaosun]

Unable to compile mamba under Windows:

Is there any difference between the mamba in this repository and the official one?
https://github.com/state-spaces/mamba

FlashDepth:

# install our local mamba package
cd mamba
export MAMBA_FORCE_BUILD=TRUE
python -m pip install --no-build-isolation .

Building wheels for collected packages: mamba_ssm
Building wheel for mamba_ssm (pyproject.toml) ... error

Because the compilation failed, I used the latest official mamba_ssm==2.2.4 version, but it gave an error when I ran it.

pip show mamba_ssm
Name: mamba_ssm
Version: 2.2.4

flashdepth\mamba.py", line , in forward
[rank0]:     x = self.mamba(x, inference_params=inference_params) # x = torch.Size([1, 112, 64])

lib\site-packages\mamba_ssm\modules\mamba2.py", line, in step
[rank0]:     assert hidden_states.shape[1] == 1, "Only support decoding with 1 token at a time for now"
[rank0]: AssertionError: Only support decoding with 1 token at a time for now

There are too many errors, maybe the versions don't match... ?

[gene-chou]

We modified the mamba2.py file. Specifically, we decode multiple tokens at once during inference, so you're getting the error "Only support decoding with 1 token at a time for now" from the original code.

You can install from the original repo first, then manually copy over the file mamba/mamba_ssm/modules/mamba2.py from our repo to your environment (e.g. $conda_env_path/.../site-packages/mamba-ssm/modules/mamba2.py). Let me know if this works.