Can't see logger output

[parmarsuraj99]

🐛 Bug

Information

Model I am using (Bert, XLNet ...): RoBERTa

Language I am using the model on (English, Chinese ...): Sanskrit

The problem arises when using:

• the official example scripts: (give details below)
• my own modified scripts: (give details below)

The tasks I am working on is:

• an official GLUE/SQUaD task: (give the name)
• my own task or dataset: (give details below)

To reproduce

Steps to reproduce the behavior:

Can't see logger output showing model config and other parameters in Trainer that were printed in training_scripts.

from transformers import Trainer, TrainingArguments

training_args = TrainingArguments(
    output_dir="./model_path",
    overwrite_output_dir=True,
    num_train_epochs=1,
    per_gpu_train_batch_size=128,
    per_gpu_eval_batch_size =256,
    save_steps=1_000,
    save_total_limit=2,
    logging_first_step = True,
    do_train=True,
    do_eval = True,
    evaluate_during_training=True,
    logging_steps = 1000
)

trainer = Trainer(
    model=model,
    args=training_args,
    data_collator=data_collator,
    train_dataset=train_dataset,
    eval_dataset = valid_dataset,
    prediction_loss_only=True,
)

%%time
trainer.train(model_path="./model_path")

Is it It is overriden by tqdm?
but I can still see Using deprecated --per_gpu_train_batch_sizeargument which will be removed in a future version. Using--per_device_train_batch_size is preferred.

Environment info

• transformers version: 2.10.0
• Platform: Linux-4.19.104+-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.6.9
• PyTorch version (GPU?): 1.6.0a0+916084d (False)
• Tensorflow version (GPU?): not installed (NA)
• Using GPU in script?: TPU
• Using distributed or parallel set-up in script?: No

[LysandreJik]

Hi, have you tried setting the logging level to INFO? You can do so with the following lines:

import logging

logging.basicConfig(level=logging.INFO)

[parmarsuraj99]

It worked! Thanks

[jawadSajid]

Hey, this doesn't log the training progress by trainer.train() into a log file. I want to keep appending the training progress to my log file but all I get are the prints and the parameters info at the end of trainer.train(). What would be a way around to achieve this? @parmarsuraj99 @LysandreJik

[iamlockelightning]

+1

same request. @parmarsuraj99 @LysandreJik

[iamlockelightning]

Share a solution, not so elegant but works.

I define a new Callback function, which logging the logs using the outside logger. And then pass it to the trainer.

class LoggerLogCallback(transformers.TrainerCallback):
    def on_log(self, args, state, control, logs=None, **kwargs):
        control.should_log = False
        _ = logs.pop("total_flos", None)
        if state.is_local_process_zero:
            logger.info(logs) # using your custom logger