search

cure-lab / MagicDrive

[ICLR24] Official implementation of the paper “MagicDrive: Street View Generation with Diverse 3D Geometry Control”
https://gaoruiyuan.com/magicdrive/
GNU Affero General Public License v3.0
419 stars 22 forks source link

No module named 'xformers._flash_attn' #47

Closed haichen-ber closed 4 days ago

haichen-ber commented 5 days ago

Hello, thanks for your excellent work, I am getting an error ""No module named 'xformers._flash_attn'"" when I run the demo command you gave. Environment: CUDA10.2 PyTorch 1.10.2 gcc 7.4.0 NVIDIA A100

flymin commented 5 days ago

I am not sure about the reason. Could you try https://github.com/cure-lab/MagicDrive/tree/disable-flash and reinstall the xformers? If not solved, please provide more trace.

haichen-ber commented 5 days ago

I am not sure about the reason. Could you try https://github.com/cure-lab/MagicDrive/tree/disable-flash and reinstall the xformers? If not solved, please provide more trace.

I've tried that and it still reports the same error. More detailed error reporting is: 531bf090bab9d885698287d3d2d6d93

flymin commented 5 days ago

Did you install triton? Can you remove it and try?

haichen-ber commented 5 days ago

Did you install triton? Can you remove it and try?

I tried it and it still reports the same error. I even switched to a different version of triton.

flymin commented 5 days ago

Let me explain.

We never installed triton and the environment without it should be fine. I think you may need to remove triton and rebuild xformers again. However, for a quick workaround, please try to add XFORMERS_FORCE_DISABLE_TRITON=1 to your env. Anyway, this function should return False. You may take a closer look.

haichen-ber commented 5 days ago

Let me explain.

We never installed triton and the environment without it should be fine. I think you may need to remove triton and rebuild xformers again. However, for a quick workaround, please try to add XFORMERS_FORCE_DISABLE_TRITON=1 to your env. Anyway, this function should return False. You may take a closer look.

Ok, I re-downloaded xformers based on your comments, but I get the following error: ![Uploading 24af7786a1ce6d5741a1d53f17813e0.png…]()

flymin commented 5 days ago

Your figure is not working. BTW, please make sure you install xformers in this repo.

haichen-ber commented 5 days ago

Your figure is not working. BTW, please make sure you install xformers in this repo.

I'm sure I downloaded xformers from this repo, but it really doesn't work for me. 24af7786a1ce6d5741a1d53f17813e0

flymin commented 5 days ago

Ok, now it is clearly another problem.

Did you download and prepare all the pre-trained weights as described in the readme? By default, nothing would be downloaded automatically.

BTW, although it is not related to your problem above, I note that you are using A100 with CUDA10.2, which is not compatible. We have docs for the A-series card.

haichen-ber commented 4 days ago

Ok, now it is clearly another problem.

Did you download and prepare all the pre-trained weights as described in the readme? By default, nothing would be downloaded automatically.

BTW, although it is not related to your problem above, I note that you are using A100 with CUDA10.2, which is not compatible. We have docs for the A-series card.

Ok, now it is clearly another problem.

Did you download and prepare all the pre-trained weights as described in the readme? By default, nothing would be downloaded automatically.

BTW, although it is not related to your problem above, I note that you are using A100 with CUDA10.2, which is not compatible. We have docs for the A-series card.

Thanks, it's solved！！！！！！！！！！！！！！！！！