No Module Named 'torch'

[MilesQLi]

When I run pip install flash-attn, it says that. But obviously, it is wrong. See screenshot.

[jackbravo]

anyone having problems with macbookpro m3?

@ManuelSokolov. Seeing the requirements of the README, it says that this project requires CUDA, so I don't think you can run it with an M3. But also, on the file usage.md it mentions a fork for apple silicon: https://github.com/philipturner/metal-flash-attention

[Calandiel]

The proposed solutions here are (imho) problematic. Using an nvidia container is a huge dependency and I personally find it difficult to trust a build with no isolation (to not mess up my existing setup) for a library that seems to have a messy installation process.

[albertotono]

Dear Professor @tridao and team,

I had the same issue with these

• flash-attn==2.5.6,
• causal-conv1d==1.1.3.post1
• mamba-ssm==1.1.4

I have torch = 2.1 and CUDA 12.2, I am following this https://github.com/kuleshov-group/mdlm

In my case I solve add in the .tomldirectly the wheel

• mamba-ssm = {url = "https://github.com/state-spaces/mamba/releases/download/v2.2.2/mamba_ssm-2.2.2+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl"}
• flash-attn = {url = "https://github.com/Dao-AILab/flash-attention/releases/download/v2.5.6/flash_attn-2.5.6+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl"}
• causal-conv1d = {url = "https://github.com/Dao-AILab/causal-conv1d/releases/download/v1.1.3.post1/causal_conv1d-1.1.3.post1+cu122torch2.1cxx11abiTRUE-cp310-cp310-linux_x86_64.whl"}

and also using

sudo poetry add "mamba-ssm @ https://github.com/state-spaces/mamba/releases/download/v2.2.2/mamba_ssm-2.2.2+cu122torch2.1cxx11abiFALSE-cp310-cp310-linux_x86_64.whl"

sudo poetry add "causal-conv1d  @ https://github.com/Dao-AILab/causal-conv1d/releases/download/v1.1.3.post1/causal_conv1d-1.1.3.post1+cu122torch2.1cxx11abiTRUE-cp310-cp310-linux_x86_64.whl"

I hope it helps for all the community.