Skip to content

Make pp_stream wait on attn_backward_dx #10984

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: dsv3_dev
Choose a base branch
from
Open

Make pp_stream wait on attn_backward_dx #10984

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
21 changes: 13 additions & 8 deletions paddlenlp/transformers/deepseek_v2/modeling_pp.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1077,7 +1077,7 @@ def forward_backward(self, inputs, output_grad, combine_bw_event_to_wait=None, p
# get dispatch backward event
dispatch_backward_event = deep_ep.get_event_from_comm_stream(self.backward_node.moe_group.id)

paddle.base.core.nvprof_nvtx_push("dispatch_backward_dw")
paddle.base.core.nvprof_nvtx_push("mlp_backward_dw")
WeightGradStore.pop()
assert WeightGradStore.funcs_queue.empty()
paddle.base.core.nvprof_nvtx_pop()
Expand Down Expand Up @@ -1108,11 +1108,7 @@ def forward_backward(self, inputs, output_grad, combine_bw_event_to_wait=None, p

if pp_stream is not None:
send_recv_stream = paddle.device.Stream(stream_base=pp_stream)

# combine_forward_event.custom_stream_wait( pp_stream)
# final_out_event.custom_stream_wait(pp_stream)

paddle.base.core.nvprof_nvtx_push("pp stream add")
paddle.base.core.nvprof_nvtx_push("pp_stream_add")

with paddle.device.stream_guard(send_recv_stream):
combine_forward_event.current_stream_wait()
Expand All @@ -1127,20 +1123,27 @@ def forward_backward(self, inputs, output_grad, combine_bw_event_to_wait=None, p

dispatch_backward_event.calc_stream_wait(self.backward_node.moe_group.id)

paddle.base.core.nvprof_nvtx_push("attn_backward")
paddle.base.core.nvprof_nvtx_push("attn_backward_dx")
assert WeightGradStore.funcs_queue.empty()
WeightGradStore.enabled = True
output_grad = self.backward_node.attn_backward(output_grad)
event_to_wait = deep_ep.get_event_from_calc_stream(self.backward_node.moe_group.id)
paddle.base.core.nvprof_nvtx_pop()

if EventStore is not None:
EventStore.set(event_to_wait)

if pp_stream is not None:
# TODO(liangshuhao): this wait seems unnecessary, but there will be
# convergence issue without this.
with paddle.device.stream_guard(send_recv_stream):
event_to_wait.current_stream_wait()

paddle.base.core.nvprof_nvtx_push("attn_backward_dw")
WeightGradStore.enabled = False
WeightGradStore.flush()
WeightGradStore.pop()
assert WeightGradStore.funcs_queue.empty()

paddle.base.core.nvprof_nvtx_pop()

# residual add
Expand Down Expand Up @@ -1218,6 +1221,8 @@ def forward_backward(self, inputs, output_grad, combine_bw_event_to_wait=None, p
paddle.base.core.nvprof_nvtx_pop() # moe_attn

paddle.base.core.nvprof_nvtx_push("dense_mlp_moe_dispatch")
if combine_bw_event_to_wait is not None:
combine_bw_event_to_wait.calc_stream_wait(self.forward_node.moe_group.id)
output_grad = self.backward_node.mlp_node.backward(output_grad)
inputs = self.forward_node.dispatch_forward(
inputs, previous_event=attn_fw_event, async_finish=True, allocate_on_comm_stream=True
Expand Down
Loading