Skip to content
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

Jump to bottom

Move LR Step #6575

Closed
wants to merge 2 commits into from
Closed

Move LR Step #6575

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1d638cd
lr step should go before optimizer step for proper updates
jomayeri Sep 26, 2024
1602ad0
Merge branch 'master' into jomayeri/lr-step-move
jomayeri Sep 27, 2024
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
21 changes: 10 additions & 11 deletions deepspeed/runtime/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,7 +225,7 @@ def __init__(self,
self.num_experts = []
self.gate_modules = []
self.moe_layers = []
self._step_applied = False
self._step_applied = True
self._global_grad_norm = None
self.use_ds_comm = False # False --> Use torch.dist, True --> Use ds.comm backend.

Expand Down Expand Up @@ -2108,6 +2108,15 @@ def clip_fp32_gradients(self):
clip_grad_norm_(parameters=self.module.parameters(), max_norm=self.gradient_clipping(), mpu=self.mpu)

def _take_model_step(self, lr_kwargs, block_eigenvalue={}):
if self.lr_scheduler is not None and self._step_applied:
try:
self.lr_scheduler.step(**(lr_kwargs or {}))
except TypeError:
# XXX Hack to work with Megatron 2.0 and DeepSpeed pipelines.
# We don't currently have a way to specify lr_kwargs from
# pipe_engine.train_batch()
self.lr_scheduler.step(self.train_batch_size())

if self.gradient_clipping() > 0.0:
if not (self.fp16_enabled() or self.bfloat16_enabled() or self.amp_enabled() or self.zero_optimization()):
self.clip_fp32_gradients()
Expand Down Expand Up @@ -2157,14 +2166,6 @@ def _take_model_step(self, lr_kwargs, block_eigenvalue={}):
self.skipped_steps += 1
else:
self.compression_scheduler.step()
if self.lr_scheduler is not None:
try:
self.lr_scheduler.step(**(lr_kwargs or {}))
except TypeError:
# XXX Hack to work with Megatron 2.0 and DeepSpeed pipelines.
# We don't currently have a way to specify lr_kwargs from
# pipe_engine.train_batch()
self.lr_scheduler.step(self.train_batch_size())

if report_progress and (self.global_steps + 1) % self.steps_per_print() == 0:
self._report_progress(self.global_steps + 1)
Expand All @@ -2191,8 +2192,6 @@ def step(self, lr_kwargs=None):

report_progress = False

self._step_applied = False # assume False, will flip to True

# Update the model when we reach gradient accumulation boundaries
if self.is_gradient_accumulation_boundary():
self.gas_boundary_ctr += 1
Expand Down
Loading