Skip to content

Support multimodule communication #2031

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

Merged
yaoyu-33 merged 23 commits into from
Jan 24, 2026
Merged

Support multimodule communication #2031

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
23 commits
Select commit Hold shift + click to select a range
5aec371
Support multimodule communication
shifangx Oct 5, 2025
01f2b4f
add test
shifangx Oct 6, 2025
3425fac
add test for MultiModulePipelineCommunicator init
shifangx Oct 6, 2025
cc2b6ee
add test for compute_total_pipeline_stages
shifangx Oct 6, 2025
e940fcb
add test_send_forward_recv_forward
shifangx Oct 6, 2025
b689f16
add test_send_backward_recv_backward
shifangx Oct 6, 2025
9565a98
add test_send_forward_recv_backward_send_backward_recv_forward
shifangx Oct 6, 2025
381986a
skip if torch version is low
shifangx Oct 6, 2025
f064f05
delete duplicated dist.broadcast in BridgeCommunicator
shifangx Oct 6, 2025
11ae642
add test_send_forward_recv_forward_with_transformer_blocks
shifangx Oct 7, 2025
698e454
add test_send_forward_recv_forward_with_transformer_blocks_and_differ…
shifangx Oct 7, 2025
babbb54
chore: Format files
Oct 7, 2025
2ac3216
update comment and logging
shifangx Oct 14, 2025
e9d03ba
rename from pp_stage to pp_rank, and import some unit function from t…
shifangx Oct 14, 2025
7c10cb0
add test for the case when pp(image_encoder) != pp(audio_encoder)
shifangx Oct 14, 2025
b28287f
chore: Format files
Oct 14, 2025
fbeb6e9
add test with llm pp size larger than 2, such as llm pp=4
shifangx Oct 22, 2025
3b52ec4
move comments to docstrings
yaoyu-33 Nov 13, 2025
aa19652
Merge branch 'main' into shifang/multi_module_comm
yaoyu-33 Dec 10, 2025
10bf0c4
add copyright
shifangx Jan 6, 2026
19f0ecd
Merge branch 'main' into shifang/multi_module_comm
shifangx Jan 6, 2026
2fe2eeb
Merge branch 'main' into shifang/multi_module_comm
yaoyu-33 Jan 16, 2026
762ec65
Merge branch 'main' into shifang/multi_module_comm
shifangx Jan 24, 2026
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
3 changes: 0 additions & 3 deletions megatron/core/pipeline_parallel/bridge_communicator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -628,9 +628,6 @@ def send_forward_recv_backward(
dist.broadcast(
shape_tensor, src=self.current_rank, group=self.src_grid_broadcast_pg
)
dist.broadcast(
shape_tensor, src=self.current_rank, group=self.src_grid_broadcast_pg
)

# Broadcast the tensors to all ranks in the group
dist.broadcast(
Expand Down
Loading
Loading