cannnot install/import flash_attn

[yukyeongmin]

python 3.8.16
cuda 11.3
torch 1.12.0
gpu geforce rtx 3090

with flash-attn 0.2
note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed building wheel for flash-attn Running setup.py clean for flash-attn Failed to build flash-attn ERROR: Could not build wheels for flash-attn, which is required to install pyproject.toml-based projects
failed to build when i tried pip install flash-attn==0.2

so i tried with newest version
with flash-attn 2.5.6
Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/home/cvnar1/anaconda3/envs/VideoComposer/lib/python3.8/site-packages/flash_attn/__init__.py", line 3, in <module> from flash_attn.flash_attn_interface import ( File "/home/cvnar1/anaconda3/envs/VideoComposer/lib/python3.8/site-packages/flash_attn/flash_attn_interface.py", line 10, in <module> import flash_attn_2_cuda as flash_attn_cuda ImportError: /home/cvnar1/anaconda3/envs/VideoComposer/lib/python3.8/site-packages/flash_attn_2_cuda.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN3c104impl8GPUTrace13gpuTraceStateE

please let me know why this error is occured.

[cmh1027]

try newest version of flash_attn and change mha_flash.py to use "from flash_attn import flash_attn_qkvpacked_func"