Environment about flash attention

[WeixuanQiao]

Thanks for your excellent Work in this area firstly. When I tried to inference the model, the error as following:

 File "/mnt/HDD/home_combine/qwx/Prompt4Driving/projects/mmdet3d_plugin/models/utils/attention.py", line 16, in <module>
    from flash_attn.flash_attn_interface import flash_attn_unpadded_kvpacked_func
ModuleNotFoundError: No module named 'flash_attn'

flash_attn need CUDA>=11.7, but the project's CUDA ==11.1.

Collecting flash_attn
  Using cached flash_attn-2.7.0.post2.tar.gz (2.7 MB)
  Preparing metadata (setup.py) ... error
  error: subprocess-exited-with-error
  
  × python setup.py egg_info did not run successfully.
  │ exit code: 1
  ╰─> [12 lines of output]
      fatal: not a git repository (or any of the parent directories): .git
      Traceback (most recent call last):
        File "<string>", line 2, in <module>
        File "<pip-setuptools-caller>", line 34, in <module>
        File "/tmp/pip-install-r4uz0u_i/flash-attn_32a89c6d2cd64a458e47c873c4dc9a8a/setup.py", line 164, in <module>
          raise RuntimeError(
      RuntimeError: FlashAttention is only supported on CUDA 11.7 and above.  Note: make sure nvcc has a supported version by running nvcc -V.
      
      
      torch.__version__  = 1.9.0+cu111
      
      
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

Could you tell me, how to deal with this problem?

[wudongming97]

We used a very early version of flash-attention: https://github.com/Dao-AILab/flash-attention/tree/v0.2.2

You can try to install it locally.

[WeixuanQiao]

Thank you for your prompt reply. I will try it out

[djandjac]

请问您解决了吗 我也出现了这个问题