modelzoo.common.pytorch.run_cstorch_flow#
Generic run scripts build using the cstorch API
Functions
Compute the model wise and per layer norm of the gradients |
|
Compute the model wise norm of the parameters |
|
Get the path to the checkpoint with the highest global step |
|
Get the path to the model checkpoint, if any. |
|
Log the input tensors to tensorboard |
|
Customized equivalent to cstorch.amp.optimizer_step additionally featuring grad norm summaries |
|
Runs the evaluatiion workflow built using the cstorch API |
|
Set up the cstorch run and call the appropriate helper based on the mode |
|
Runs the training workflow built using the cstorch API |
Classes
Dataclass for parsing grad scaler params from optimizer params. |