FedScale
FedScale copied to clipboard
SymbioticLab
Unable to Train Google Speech Dataset
What happened + What you expected to happen
We are unable to train on the google speech dataset. While training, we get the error as shown below Please help us out.
Versions / Dependencies
We are running on Red Hat servers and are using python 3.7
Reproduction script
It seems like the problem is with resnet code for it is error due to mismatch in the number of channels.
Issue Severity
High: It blocks me from completing my task.
Thanks for trying FedScale. Note that the model you are using perhaps is from torchvision. Please refer to these specialized models for speech task: https://github.com/SymbioticLab/FedScale/blob/master/fedscale/cloud/fllibs.py#L129-L162
Thanks for the reply @fanlai0990 . However we are using the same model as provided in the repository. Also we are using the same configurations as provided in the config file: https://github.com/SymbioticLab/FedScale/blob/master/benchmark/configs/speech/google_speech.yml Especially the task which is set to speech.
Hello, I'm encountering the same problem mentioned by @implosion07 . I've used the provided model from the repository and followed the configurations exactly as specified in the config file https://github.com/SymbioticLab/FedScale/blob/master/benchmark/configs/speech/google_speech.yml . Unfortunately, I haven't been able to resolve the issue. Just wanted to add my experience here to raise awareness about the problem. Thanks in advance!
Thanks for letting us know. I just finished my test run, but it seems the code works well for me. Did you install FedScale from the source code?
Thanks for checking @fanlai0990 We are using the following configuration. Can you help us in finding out the mistake if any.
It's hard for us to identify the issue as we can not reproduce it. For now, (1) it seems the data_dir is slightly different to the default one. Please double check it; (2) roll back to the default configuration here to check it.