Learning rate schedules and Mamba layer #282
Merged
Conversation
This file contains hidden or 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
specify full path to checkpoint using --load-cehckpoint
erwulff
changed the title
farakiko
pushed a commit
to farakiko/particleflow
that referenced
this pull request
Jan 23, 2024
* fix: update parameter files * fix: better comet-ml logging * update flatiron Ray Train submissions scripts * update sbatch script * log overridden config to comet-ml instead of original * fix: checkpoint loading specify full path to checkpoint using --load-cehckpoint * feat: implement LR schedules in the PyTorch training code * update sbatch scripts * feat: LR schedules support checkpointing and resuming training * update sbatch scripts * update ray tune search space * fix: dropout parameter not taking effect on torch gnn-lsh model * make more gnn-lsh parameters confgiurable * make activation function configurable * update raytune search space * feat: add MambaLayer * update raytune search space * update pyg-cms.yaml * fix loading of checkpoint in testing with raytrain based run
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
Implement LR schedules in the PyTorch training code. Add MambaLayer and make it configurable through parameter config files and for HPO.
Also contains the following changes.
--loadnow takes as input the path to a saved checkpoint from which to start a new training. I.e., a training starts from epoch 1, with LR schedules starting from the beginning. The only thing that is loaded is the pre-trained model weights from the checkpoint.--resume-trainingis a new command line arg that takes as input a path to a training directory containing an unfinished training and attempts to restore it and continue the training from the last saved checkpoint.Learning rate versus training step of a training using the
cosinedecayLR schedule and was interrupted halfway through and then continued.Learning rate versus training step of a training using the
onecycleLR schedule and was interrupted halfway through and then continued.