You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Context:
I am trying to train a Yolo-NAS model and store my checkpoints remotely on an AWS s3 bucket.
Among others, I pass the following training parameters to the Trainer:
...but to no effect. The checkpoints don't appear on my s3 bucket.
The culprit seems to be these few lines in the Trainer class definition (I removed some of the code for clarity):
def_initialize_sg_logger_objects(self, additional_configs_to_log: Dict=None):
"""Initialize object that collect, write to disk, monitor and store remotely all training outputs"""sg_logger=core_utils.get_param(self.training_params, "sg_logger")
# OVERRIDE SOME PARAMETERS TO MAKE SURE THEY MATCH THE TRAINING PARAMETERSgeneral_sg_logger_params= {
"experiment_name": self.experiment_name,
"storage_location": "local",
"resumed": self.load_checkpoint,
"training_params": self.training_params,
"checkpoints_dir_path": self.checkpoints_dir_path,
}
ifisinstance(sg_logger, str):
sg_logger_cls=SG_LOGGERS.get(sg_logger)
sg_logger_params=core_utils.get_param(self.training_params, "sg_logger_params", {})
ifissubclass(sg_logger_cls, BaseSGLogger):
sg_logger_params= {**sg_logger_params, **general_sg_logger_params}
It seems that whatever the user specifies, the storage_location parameter of the logger will be overriden to "local" whenever the logger class inherits from BaseSGLogger.
My question.
Why is this? And how do I setup my logger so it stores checkpoints and logs to s3?
Versions
No response
The text was updated successfully, but these errors were encountered:
馃挕 Your Question
Hi there,
Context:
I am trying to train a Yolo-NAS model and store my checkpoints remotely on an AWS s3 bucket.
Among others, I pass the following training parameters to the
Trainer
:...but to no effect. The checkpoints don't appear on my s3 bucket.
The culprit seems to be these few lines in the
Trainer
class definition (I removed some of the code for clarity):It seems that whatever the user specifies, the
storage_location
parameter of the logger will be overriden to"local"
whenever the logger class inherits fromBaseSGLogger
.My question.
Why is this? And how do I setup my logger so it stores checkpoints and logs to s3?
Versions
No response
The text was updated successfully, but these errors were encountered: