Skip to content

[https://nvbugs/5667774][fix] Refine Piecewise Cuda Graph Condition for DP #9393

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
liji-nv merged 1 commit into from
Dec 2, 2025
Merged

[https://nvbugs/5667774][fix] Refine Piecewise Cuda Graph Condition for DP #9393

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
15 changes: 14 additions & 1 deletion tensorrt_llm/_torch/pyexecutor/model_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1224,6 +1224,11 @@ def _get_all_rank_num_tokens(self, attn_metadata: AttentionMetadata):
return list(self.dist.tp_allgather(attn_metadata.num_tokens))
return None

def _get_all_rank_ctx_requests(self, num_ctx_requests: int):
if self.enable_attention_dp:
return list(self.dist.tp_allgather(num_ctx_requests))
return None

def _get_padding_params(
self, total_num_tokens: int, num_ctx_requests: int,
attn_all_rank_num_tokens: Optional[List[int]]
Expand All @@ -1237,6 +1242,9 @@ def _get_padding_params(
"""
padded_num_tokens = total_num_tokens

all_rank_ctx_requests = self._get_all_rank_ctx_requests(
num_ctx_requests)

def get_padded_piecewise_tokens(tokens):
captured_num_tokens = self._torch_compile_backend.capture_num_tokens
return captured_num_tokens[bisect.bisect_left(
Expand All @@ -1249,7 +1257,12 @@ def get_padded_piecewise_tokens(tokens):
-1]
# Torch piecewise cuda graph is enabled.
if attn_all_rank_num_tokens is not None:
can_run_piecewise_cuda_graph = (num_ctx_requests != 0 and
# Any rank has context requests, we enable piecewise cuda graph.
has_ctx_requests = num_ctx_requests != 0 or (
all_rank_ctx_requests is not None
and any(ctx_requests != 0
for ctx_requests in all_rank_ctx_requests))
can_run_piecewise_cuda_graph = (has_ctx_requests and
max(attn_all_rank_num_tokens)
<= max_captured_num_tokens)
all_ranks_can_run_piecewise_cuda_graph = list(
Expand Down