Skip to content

Refine prefill CUDA graph capture sizes #1331

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
shihaobai merged 2 commits into from
Jun 4, 2026
Merged

Refine prefill CUDA graph capture sizes #1331

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
af897dd
prefill cudagraph capture size
shihaobai Jun 4, 2026
76ee67b
fix
shihaobai Jun 4, 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
16 changes: 10 additions & 6 deletions lightllm/common/basemodel/prefill_cuda_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,14 +31,18 @@ def __init__(self, decode_cuda_graph: CudaGraph, tp_world_size: int):
self.enable_prefill_microbatch_overlap = self.args.enable_prefill_microbatch_overlap
self.max_handle_token_num = self.args.prefill_cudagraph_max_handle_token

graph_handle_token_nums = []
for i in range(2048):
token_num = int(2 ** (2 * i))
if 1 < token_num < self.max_handle_token_num:
graph_handle_token_nums.append(token_num)
graph_handle_token_nums = (
list(range(4, 33, 4))
+ list(range(48, 257, 16))
+ list(range(288, 513, 32))
+ list(range(576, 1024 + 1, 64))
+ list(range(1280, 4096 + 1, 256))
+ list(range(4608, self.max_handle_token_num + 1, 512))
)
graph_handle_token_nums = [e for e in graph_handle_token_nums if e <= self.max_handle_token_num]
graph_handle_token_nums.append(self.max_handle_token_num)

graph_handle_token_nums = list(set(graph_handle_token_nums))
graph_handle_token_nums = list(set[int](graph_handle_token_nums))

@gemini-code-assist gemini-code-assist Bot Jun 4, 2026

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

critical

Using set[int](...) at runtime will raise a TypeError: 'types.GenericAlias' object is not callable in Python. Subscription of built-in types like set[...] is only for type annotations and cannot be instantiated directly. Use set(...) instead.

Suggested change
graph_handle_token_nums = list(set[int](graph_handle_token_nums))
graph_handle_token_nums = list(set(graph_handle_token_nums))

graph_handle_token_nums.sort()
if self.args.enable_tpsp_mix_mode:
graph_handle_token_nums = [
Expand Down
Loading