Skip to content
New issue

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

#

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

Already on GitHub? # to your account

[Distributed] [Cherry-Pick] support fuse optimizer (#9519) #9777

Merged
merged 1 commit into from
Jan 14, 2025
Merged

[Distributed] [Cherry-Pick] support fuse optimizer (#9519) #9777

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
9 changes: 7 additions & 2 deletions paddlenlp/trainer/training_args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -293,6 +293,7 @@ class TrainingArguments:
enable_stage1_allgather_overlap, overlap stage1 V2 allgather with next step forward computation. There are some constraints for the overlap, such as the logging_step should be bigger than 1 for allgather overlap forward compute and no other sync could be called during the training for allgather overlap.
disable_stage1_reduce_avg, replace reduce_avg with original reduce_sum+scale in stage1, which can be used for accuracy verification.
enable_release_grads, reduce peak memory usage by releasing gradients after each iteration. The creation of gradients will be postponed until backward propagation of the next iteration.
enable_fuse_optimizer_states, fuse optimizer states to a single storage.
recompute (`bool`, *optional*, defaults to `False`):
Recompute the forward pass to calculate gradients. Used for saving memory.
Only support for networks with transformer blocks.
Expand Down Expand Up @@ -1412,10 +1413,11 @@ def is_segment_parallel_supported():
"enable_stage1_broadcast_overlap",
"enable_stage1_allgather_overlap",
"enable_release_grads",
"enable_fuse_optimizer_states",
]:
raise ValueError(
f"Found unknown pipeline mode config {x}, "
f"accpet config is enable_stage1_tensor_fusion, enable_stage1_overlap, enable_stage2_overlap, split_param, disable_stage1_reduce_avg, enable_stage1_broadcast_overlap, enable_stage1_allgather_overlap."
f"Found unknown sharding mode config {x}, "
f"accpet config is enable_stage1_tensor_fusion, enable_stage1_overlap, enable_stage2_overlap, split_param, disable_stage1_reduce_avg, enable_stage1_broadcast_overlap, enable_stage1_allgather_overlap, enable_release_grads, enable_fuse_optimizer_states."
)
if "disable_stage1_reduce_avg" in sharding_parallel_config:
assert self.sharding == [
Expand All @@ -1441,6 +1443,9 @@ def is_segment_parallel_supported():
if "enable_release_grads" in sharding_parallel_config:
strategy.hybrid_configs["sharding_configs"].release_gradients = True

if "enable_fuse_optimizer_states" in sharding_parallel_config:
strategy.hybrid_configs["sharding_configs"].enable_fuse_optimizer_states = True

if self.pipeline_parallel_degree == 1:
strategy.hybrid_configs["sharding_configs"].tensor_fusion = (
True if "enable_stage1_tensor_fusion" in sharding_parallel_config else False
Expand Down