Tong Chen

Hi, could you please release some pretrained model or some command about the training and testing? Besides that, could you please draw a dataset format tree to let readers have...

Tensor shape is mis-matched

2

clip_caps[i] = clip_text.mean(0) ~~~~~~~~~^^^ RuntimeError: The expanded size of the tensor (59136) must match the existing size (768) at non-singleton dimension 0. Target sizes: [59136]. Tensor sizes: [768] 

Thank you for your impactable work! I am really interested in this work and I am really interested in how you preprocessed the data. Could you release the code about...

``` [rank1]:[E626 06:24:44.903881913 ProcessGroupNCCL.cpp:616] [Rank 1] Watchdog caught collective operation timeout: WorkNCCL(SeqNum=808, OpType=ALLREDUCE, NumelIn=9801523, NumelOut=9801523, Timeout(ms)=600000) ran for 600020 milliseconds before timing out. [rank1]:[E626 06:24:44.906457629 ProcessGroupNCCL.cpp:1785] [PG ID 0 PG...

我尝试去加载stage1的模型权重作为第二个模型的一部分去进行训练，但是发现权重会被load进错误的模块中。我两个模型的权重命名是正确的，请问这样的编写是正确的吗。