Some other settings that are unrelated to the specific model structure or hyperparameters. Such as logging, checkpoint saving, etc.
🤗 In our codebase, many logging methods are integrated, including tensorboard/wandb/etc.
In our default scripts, we set use_wandb
to False
to disable wandb logging, because it requires creating an account and making some additional settings, which increases the user's workload. However, if you believe you need to enable wandb logging (which I think is more elegant), you will need to set it up as follows:
- add
--use-wandb True
to the script. - set the
EXP_OWNER
or--exp-owner
, which is the wandb account name. - manually record your current git revision number (
--git-version
, for example,--git-version d293ceee8c6d208bb4a5d7b6ba92a7b5d7ec4bca
) to ensure you can rollback and reproduce the experiment. I know this is not very elegant, but it is easy enough 😛.