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

[Dygraph] Support skip_check_meta in eval mode of Pipeline #63001

Merged
merged 1 commit into from
Mar 26, 2024
Merged

[Dygraph] Support skip_check_meta in eval mode of Pipeline #63001

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
8 changes: 6 additions & 2 deletions python/paddle/distributed/fleet/meta_parallel/pipeline_parallel.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -727,7 +727,9 @@ def eval_batch(self, data, compute_loss=False):

output_tensor = self._forward_step(input_tensor, micro_dataset)
self._p2p_helper.send_forward(
output_tensor, self.is_pipeline_last_stage()
output_tensor,
self.is_pipeline_last_stage(),
skip_check_meta=True,
)

input_buffers.append(input_tensor)
Expand All @@ -743,7 +745,9 @@ def eval_batch(self, data, compute_loss=False):

output_tensor = self._forward_step(input_tensor, micro_dataset)
self._p2p_helper.send_forward(
output_tensor, self.is_pipeline_last_stage()
output_tensor,
self.is_pipeline_last_stage(),
skip_check_meta=True,
)

input_buffers.append(input_tensor)
Expand Down
6 changes: 3 additions & 3 deletions python/paddle/distributed/fleet/meta_parallel/pp_utils/four_directions_p2p_communication.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -692,7 +692,7 @@ def __init__(self, use_cache=True):
self._send_recv_meta = SendRecvMeta()
self._use_cache = use_cache

def _send_meta(self, output_tensor):
def _send_meta(self, output_tensor, skip_check_meta=False):
if not self._send_recv_meta.has_send_meta:
self._send_recv_meta.set_send_message(output_tensor)
self._send_recv_meta.send_meta(
Expand Down Expand Up @@ -745,12 +745,12 @@ def recv_backward(self, pp_last_stage, sync_recv=True):
_timers("recv_backward").stop()
return output_tensor_grad

def send_forward(self, output_tensor, pp_last_stage):
def send_forward(self, output_tensor, pp_last_stage, skip_check_meta=False):
global _timers
if _timers is not None:
_timers("send_forward").start()
if not pp_last_stage:
self._send_meta(output_tensor)
self._send_meta(output_tensor, skip_check_meta=skip_check_meta)

_p2p_helper(
tensor_send_next=output_tensor,
Expand Down
14 changes: 10 additions & 4 deletions python/paddle/distributed/fleet/meta_parallel/pp_utils/p2p_communication.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -649,14 +649,14 @@ def __init__(self, use_cache=True):
self._send_recv_meta = SendRecvMeta()
self._use_cache = use_cache

def _send_meta(self, output_tensor):
def _send_meta(self, output_tensor, skip_check_meta=False):
if not self._send_recv_meta.has_send_meta:
self._send_recv_meta.set_send_message(output_tensor)
self._send_recv_meta.send_meta(
output_tensor, _hcg.get_pipe_parallel_group()
)
self._send_recv_meta.has_send_meta = self._use_cache
else:
elif not skip_check_meta:
self._send_recv_meta.check_send_message(output_tensor)

def _recv_meta(self):
Expand Down Expand Up @@ -709,12 +709,18 @@ def recv_backward(self, pp_last_stage, sync_recv=True, batch_p2p_comm=True):
_timers("recv_backward").stop()
return output_tensor_grad

def send_forward(self, output_tensor, pp_last_stage, batch_p2p_comm=True):
def send_forward(
self,
output_tensor,
pp_last_stage,
batch_p2p_comm=True,
skip_check_meta=False,
):
global _timers
if _timers is not None:
_timers("send_forward").start()
if not pp_last_stage:
self._send_meta(output_tensor)
self._send_meta(output_tensor, skip_check_meta=skip_check_meta)

_p2p_helper(
tensor_send_next=output_tensor,
Expand Down