Skip to content

Reorder mtp_post_process after attention backward in 1F1B schedule plan #4695

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
Phlip79 merged 2 commits into from
May 16, 2026
Merged

Reorder mtp_post_process after attention backward in 1F1B schedule plan #4695

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1aded08
Reorder mtp_post_process after attn backward in 1F1B schedule plan
Apr 22, 2026
4bc4361
Merge branch 'main' into gdeng/main-pr-4430-mtp-post-process-order
gdengk May 12, 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
5 changes: 4 additions & 1 deletion megatron/core/models/common/model_chunk_schedule_plan.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,11 +241,14 @@ def run(f_layer, b_layer, f_input=None, b_grad=None, is_last_layer_in_bwd=False)
if f_layer is not None:
with f_layer.get_fp8_context():
f_input = f_layer.moe_combine.forward(f_input)
f_input = f_layer.mtp_post_process.forward(f_input)

if b_layer is not None and not b_layer.config.ep_overlap_early_attn_memory_release:
b_grad = b_layer.attn.backward(b_grad)

if f_layer is not None:
with f_layer.get_fp8_context():
f_input = f_layer.mtp_post_process.forward(f_input)

# Delay the last attn_dw in backward pass (attn_dw of the first layer)
# for overlapping with the p2p comm
if b_layer is not None and not is_last_layer_in_bwd:
Expand Down
Loading