search

OpenGVLab / InternVideo

[ECCV2024] Video Foundation Models & Data for Multimodal Understanding
Apache License 2.0
1.44k stars 88 forks source link

Error 'NoneType' object is not callable when finetuning which is located in line 272 in './InternVideo2/single_modality/models/internvideo2.py' #187

Closed Jenny199099 closed 2 months ago

Jenny199099 commented 2 months ago

Hi author, thank you for your great work. I have encountered a problem when I finetuned Internvideo. In line 272 of "./InternVideo2/single_modality/models/internvideo2.py", the function "FusedMLP" returns a NoneType value. It relates to the flash_attn module. Could you please check this part? Thank you.

yinanhe commented 2 months ago

Please ensure that the fusedmlp module of Flash_attn is correctly installed. After installing flash_attn, you need to enter the directory to compile layer_norm separately.

cd csrc/layer_norm && pip install . && cd ../../ \

See this comment for details.

Jenny199099 commented 2 months ago

Please ensure that the fusedmlp module of Flash_attn is correctly installed. After installing flash_attn, you need to enter the directory to compile layer_norm separately.

cd csrc/layer_norm && pip install . && cd ../../ \

See this comment for details.

Thank you for your reply. I have done this to compile the layer_norm module. It is successfully installed. But the problem of FusedMLP still exists.

yinanhe commented 2 months ago

Sorry for misunderstanding your point. Have you compiled and installed fused_dense_lib from this link?

Jenny199099 commented 2 months ago

Sorry for misunderstanding your point. Have you compiled and installed fused_dense_lib from this link?

It works for me. Thank you very much.

yinanhe commented 2 months ago

Thank you for your timely feedback. If you have any questions, feel free to open this issue at any time :)