Skip to content

Add ProdigyPlusScheduleFree optimizer #483

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
thrnz wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add ProdigyPlusScheduleFree optimizer #483

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion requirements.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,4 +27,5 @@ pytorch-optimizer
wandb
optimum-quanto
wheel
ninja
ninja
prodigy-plus-schedule-free
10 changes: 10 additions & 0 deletions train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -643,6 +643,9 @@ def get_optimizer(model_parameters):
elif optim_type_lower == 'genericoptim':
from optimizers import generic_optim
klass = generic_optim.GenericOptim
elif optim_type_lower == 'prodigyplusschedulefree':
from prodigyplus.prodigy_plus_schedulefree import ProdigyPlusScheduleFree
klass = ProdigyPlusScheduleFree
else:
import pytorch_optimizer
klass = getattr(pytorch_optimizer, optim_type)
Expand Down Expand Up @@ -827,20 +830,27 @@ def optimizer_hook(p):

disable_block_swap_for_eval = config.get('disable_block_swap_for_eval', False)
if config['eval_before_first_step'] and not resume_from_checkpoint:
if optimizer.__class__.__name__ == 'ProdigyPlusScheduleFree':
optimizer.eval()
evaluate(model, model_engine, eval_dataloaders, tb_writer, 0, config['eval_gradient_accumulation_steps'], disable_block_swap_for_eval)

# TODO: this is state we need to save and resume when resuming from checkpoint. It only affects logging.
epoch_loss = 0
num_steps = 0
empty_cuda_cache()
while True:
if optimizer.__class__.__name__ == 'ProdigyPlusScheduleFree':
optimizer.train()
model_engine.reset_activation_shape()
iterator = get_data_iterator_for_step(train_dataloader, model_engine)
loss = model_engine.train_batch(iterator).item()
epoch_loss += loss
num_steps += 1
train_dataloader.sync_epoch()

if optimizer.__class__.__name__ == 'ProdigyPlusScheduleFree':
optimizer.eval()

new_epoch, checkpointed, saved = saver.process_epoch(epoch, step, examples)
finished_epoch = True if new_epoch != epoch else False

Expand Down