陆秋佚

As for our setting on experiment, training epoch can not be too few, especially training from scratch, may be over 100epoch is required to ensure the result is stabel. If...

i‘am also. It seems like in during ddp trainning,when other process is backward,the main process is valid, this cause the error, but when i add torch.distributed.barry(), the error is the...

moreover when eavl_step is small ( i have tried to set it 15), it's ok when eval and back to train again.but when it's defalt, the same error always repeat...

If you have looked at the implementation of some of the upstream models in `s3prl-main/s3prl/upstream` it should be clear. such as is in `s3prl-main/s3prl/upstream/hf_hubert/expert.py` : ```python class UpstreamExpert(torch.nn.Module): def __init__(self,...

I'm just curious. In your previous experiments, did you find this to have an evident impact on performance