Skip to content

[CP] Register KV cache allgather buffer with symmetric memory #24040

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
ShangmingCai merged 3 commits into from
May 6, 2026
+9 −6
Merged

[CP] Register KV cache allgather buffer with symmetric memory #24040

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
6e7db61
[CP] Register KV cache allgather buffer with symmetric memory
wangfakang Apr 29, 2026
30b6da0
Merge branch 'main' into opt-cp
wangfakang Apr 29, 2026
7971de6
Merge branch 'main' into opt-cp
wangfakang Apr 29, 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
15 changes: 9 additions & 6 deletions python/sglang/srt/layers/utils/cp_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -179,12 +179,15 @@ def cp_all_gather_reorganized_into_tensor_kv_cache(
input_tensor = F.pad(input_tensor, padding, mode="constant", value=0)

# Create output tensor with proper shape for all dimensions
input_tensor_full = torch.empty(
max_len * cp_size,
*input_tensor.shape[1:],
device=input_tensor.device,
dtype=input_tensor.dtype,
)
with use_symmetric_memory(
get_attention_cp_group(), disabled=not is_allocation_symmetric()
):
input_tensor_full = torch.empty(
max_len * cp_size,
*input_tensor.shape[1:],
device=input_tensor.device,
dtype=input_tensor.dtype,
)

get_attention_cp_group().cp_all_gather_into_tensor_async(
input_tensor_full, input_tensor, stream
Expand Down
Loading