Skip to content

Exposing interleave argument for fused_apply_rotary_pos_emb_thd #3759

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
yaox12 merged 4 commits into from
Mar 10, 2026
Merged

Exposing interleave argument for fused_apply_rotary_pos_emb_thd #3759

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f498330
Update transformer_engine.py
huvunvidia Mar 9, 2026
8c48bd9
Add interleaved parameter to fused_apply_rotary_pos_emb_thd
huvunvidia Mar 9, 2026
efb3ebd
Update rope_utils.py
huvunvidia Mar 10, 2026
0358ee9
Merge branch 'dev' into huvu/rope_fusion_thd_interleave
huvunvidia Mar 10, 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
2 changes: 2 additions & 0 deletions megatron/core/extensions/transformer_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2385,6 +2385,7 @@ def fused_apply_rotary_pos_emb_thd(
freqs: torch.Tensor,
cp_size: int = 1,
cp_rank: int = 0,
interleaved: bool = False,
) -> torch.Tensor:
"""
Apply rotary positional embedding to input tensor T in `thd` format with CP support.
Expand All @@ -2398,6 +2399,7 @@ def fused_apply_rotary_pos_emb_thd(
cu_seqlens=cu_seqlens,
cp_size=cp_size,
cp_rank=cp_rank,
interleaved=interleaved,
)
else:
assert cp_size == 1, "Only TE >= 1.12 supports RoPE fusion for THD format with CP."
Expand Down
7 changes: 6 additions & 1 deletion megatron/core/models/common/embeddings/rope_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -313,7 +313,12 @@ def apply_rotary_pos_emb(
else:
assert fused_apply_rotary_pos_emb_thd is not None, "apply_rope_fusion is not available."
return fused_apply_rotary_pos_emb_thd(
t, cu_seqlens, freqs, cp_size=cp_group.size(), cp_rank=cp_group.rank()
t,
cu_seqlens,
freqs,
cp_size=cp_group.size(),
cp_rank=cp_group.rank(),
interleaved=config.rotary_interleaved,
)
# use unfused implementation
if cu_seqlens is None:
Expand Down
Loading