output management checkpoints and final model #630
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The aim of this PR is to provide options to separate model checkpoints from final `model.tar.gz` saved to S3. In [this](https://github.com/huggingface/notebooks/blob/master/sagemaker/08_distributed_summarization_bart_t5/sagemaker-notebook.ipynb) example, training a model should result in ~500MB, but the model saves all checkpoints which results in a 12GB `model.tar.gz` training artifact, containing all your checkpoints, most of them possibly useless for deployment. If you want to persist checkpoints to S3 while saving only the final model in the `model.tar.gz` final training artifact, you can point Hugging Face Trainer’s `output_dir` to `/opt/ml/checkpoints` and at the end of your script save a specific model to the model location with `trainer.save_model(/opt/ml/model)`. Currently, pointing Hugging Face Trainer’s `output_dir` to `/opt/ml/checkpoints` and saving location with `trainer.save_model(/opt/ml/model)` does NOT save the final model to the S3 resulting in an empty `output` folder in [this](https://github.com/huggingface/notebooks/blob/master/sagemaker/08_distributed_summarization_bart_t5/sagemaker-notebook.ipynb) example It would be useful to add `Training output management` section to this documentation after `Prepare a :hugging_face: Transformers fine-tuning script`.
|
tagging @philschmid for visibility! |
philschmid
reviewed
Jan 28, 2022
philschmid
reviewed
Jan 28, 2022
Co-authored-by: Philipp Schmid <32632186+philschmid@users.noreply.github.com>
Co-authored-by: Philipp Schmid <32632186+philschmid@users.noreply.github.com>
philschmid
reviewed
Jan 31, 2022
Co-authored-by: Philipp Schmid <32632186+philschmid@users.noreply.github.com>
replace **so** with **to** an Amazon S3 location.
mlonaws
commented
Jan 31, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The aim of this PR is to provide options to separate model checkpoints from final
model.tar.gzsaved to S3.In this example, training a model should result in ~500MB, but the model saves all checkpoints which results in a 12GB
model.tar.gztraining artifact, containing all your checkpoints, most of them possibly useless for deployment.If you want to persist checkpoints to S3 while saving only the final model in the
model.tar.gzfinal training artifact, you can point Hugging Face Trainer’soutput_dirto/opt/ml/checkpointsand at the end of your script save a specific model to the model location withtrainer.save_model(/opt/ml/model). Currently, pointing Hugging Face Trainer’soutput_dirto/opt/ml/checkpointsand saving location withtrainer.save_model(/opt/ml/model)does NOT save the final model to the S3 resulting in an emptyoutputfolder in this exampleIt would be useful to add
Training output managementsection to this documentation afterPrepare a :hugging_face: Transformers fine-tuning script.