Skip to content

Fix prefetch train pipeline #3306

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

Fix prefetch train pipeline #3306

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
13 changes: 10 additions & 3 deletions torchrec/distributed/train_pipeline/train_pipelines.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -730,6 +730,7 @@ def _pipeline_model(
batch: Optional[In],
context: TrainPipelineContext,
pipelined_forward: Type[PipelinedForward] = PipelinedForward,
custom_dist_stream: Optional[torch.Stream] = None,
) -> None:
(
self._pipelined_modules,
Expand All @@ -740,7 +741,11 @@ def _pipeline_model(
) = _rewrite_model(
model=self._model,
context=context,
dist_stream=self._data_dist_stream,
dist_stream=(
self._data_dist_stream
if custom_dist_stream is None
else custom_dist_stream
),
default_stream=torch.get_device_module(self._device).current_stream(),
batch=batch,
apply_jit=self._apply_jit,
Expand Down Expand Up @@ -768,6 +773,7 @@ def _init_pipelined_modules(
batch: In,
context: TrainPipelineContext,
pipelined_forward: Type[PipelinedForward] = PipelinedForward,
custom_dist_stream: Optional[torch.Stream] = None,
) -> None:
"""
Retrieves the pipelined modules after overriding their forwards, initializes the
Expand All @@ -779,7 +785,7 @@ def _init_pipelined_modules(
self.start_sparse_data_dist(batch, context)
return

self._pipeline_model(batch, context, pipelined_forward)
self._pipeline_model(batch, context, pipelined_forward, custom_dist_stream)

def copy_batch_to_gpu(
self,
Expand Down Expand Up @@ -1452,7 +1458,7 @@ def __init__(
else None
)
self._default_stream: Optional[torch.Stream] = (
(torch.get_device_module(self._device).Stream())
(torch.get_device_module(self._device).current_stream())
if self._device.type in ["cuda", "mtia"]
else None
)
Expand All @@ -1476,6 +1482,7 @@ def _fill_pipeline(self, dataloader_iter: Iterator[In]) -> None:
self._context,
# pyre-ignore
self._pipelined_forward_type,
self._prefetch_stream,
)
self._start_sparse_data_dist(self._batch_i)
self._wait_sparse_data_dist()
Expand Down
Loading