Li Hao
Li Hao
the version between torch and cuda unmatch, you need to use CUDA=9.0 Pytorch=1.0.1
做了几点改进: 1. 改掉了config中的bug,并精简了 2. 取消了cls_head_gcvit,统一用cls_head.py 3. 构建了metafile 我的ckpt放在/mnt/petrelfs/share_data/lihao/gcvit_ckpts/,需要帮忙上传到download.openmmlab网站
delete the cls_gcvit head
remove the SE in gcvit and use SELayer from mmcls. update the gcvit_to_mmcls.py,change the keys from fc.weight into conv.conv.weight
again solve some conflict add test_gcvit for unit testing
Thanks for Yu's reminder. We have uploaded the ecd_column_charity_new_smiles.npy in our huggingface repo: https://huggingface.co/datasets/OzymandisLi/ECDFormer_Datasets/blob/main/ECD/ecd_column_charity_new_smiles.npy Please check the file and contact me if any further questions.