Skip to content

[https://nvbugs/5451740][fix] Add DP padding back on SM120 #7965

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
peaceh-nv merged 5 commits into from
Sep 26, 2025
Merged

[https://nvbugs/5451740][fix] Add DP padding back on SM120 #7965

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
e9132c6
[https://nvbugs/5451740][fix] Add attention DP padding back on SM120
peaceh-nv Aug 11, 2025
656d3c7
Add missing module back
peaceh-nv Sep 25, 2025
f7633f7
fix comment
peaceh-nv Sep 25, 2025
52c4738
Resolve review comment
peaceh-nv Sep 25, 2025
512195c
Add a todo comment
peaceh-nv Sep 25, 2025
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
8 changes: 8 additions & 0 deletions tensorrt_llm/_torch/models/modeling_deepseekv3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -831,6 +831,14 @@ def compute_routed_output(self, hidden_states, hidden_states_fp4,
all_rank_num_tokens, do_finalize):
# max-throughput
use_dp_padding = False
# Add DP padding on SM120 for context comm performance
Comment thread
peaceh-nv marked this conversation as resolved.
# TODO: Move this model-agonostic part to MoE
if self.use_dp and self.mapping.tp_size > 1 and get_sm_version() == 120:
use_dp_padding = True
hidden_states = torch.nn.functional.pad(
hidden_states,
(0, 0, 0, max(all_rank_num_tokens) - hidden_states.shape[0]))

router_logits = self.gate(hidden_states)

routed_output = self.experts(
Expand Down