Search code examples
huggingface-transformershuggingface-tokenizersfine-tuningwandbhuggingface

wandb getting logged without initiating

I do not want to use wandb. I don't even have an account. I am simply following this notebook for finetuning. I am not running the 2nd and 3 cells because I do not want to push the model to the hub.

However, when I do trainer.train() I get the following error : enter image description here

I don't understand where wandb.log is being called. I even tried os.environ["WANDB_DISABLED"] = "true" but I still get the error. Please help.

Solution

  • posting the same message as over on transformers:

    You can turn off all external logger logging, including wandb logging by passing report_to="none" in your Seq2SeqTrainingArguments.

    You might have noticed the following warning when setting up your TrainingArguments:

    The default value for the training argument `--report_to` will change in v5 (from all installed integrations to none). In v5, you will need to use `--report_to all` to get the same behavior as now. You should start updating your code and make this info disappear :-)

    Right now the default is to run all loggers that you have installed, so maybe you installed wandb on your machine since the last time you ran the script?

    If you would like to log with wandb, best practice would already be to start setting report_to="wandb"