Problem with dropout_layer_norm

[anfortas337]

Hi, thank you for your great work! I want to use InternVideo2 as backbone. I installed flash-attn 2.7.3 and dropout_layer_norm from source. But now i am struggling with this error when trying to run sample code:

Traceback (most recent call last): File "/home/maxim/test/main.py", line 2, in import InternVideo.InternVideo2.single_modality.models.internvideo2 File "/home/maxim/test/InternVideo/InternVideo2/single_modality/models/init.py", line 1, in from .internvl_clip_vision import internvl_clip_6b File "/home/maxim/test/InternVideo/InternVideo2/single_modality/models/internvl_clip_vision.py", line 13, in from flash_attn.ops.rms_norm import DropoutAddRMSNorm File "/home/maxim/anaconda3/envs/myenv/lib/python3.12/site-packages/flash_attn/ops/rms_norm.py", line 7, in from flash_attn.ops.layer_norm import ( File "/home/maxim/anaconda3/envs/myenv/lib/python3.12/site-packages/flash_attn/ops/layer_norm.py", line 4, in import dropout_layer_norm ImportError: /home/maxim/anaconda3/envs/myenv/lib/python3.12/site-packages/dropout_layer_norm.cpython-312-x86_64-linux-gnu.so: undefined symbol: __cxa_call_terminate

Can you give any guidance how to solve it or provide config.json like you have on hugging face for VideoMae please. That will be really helpfull Thank you!

[leexinhao]

You could replace it with simple RMSnorm, see https://github.com/OpenGVLab/VideoChat-Flash/blob/657c2e7ee91b357d424ac7255b96e1809e7d5dae/llava-train_videochat/llava/model/multimodal_encoder/internvideo2/vit_scale_clean.py#L119C7-L119C14