Skip to content

Use unreg path for custom all-reduce during CUDA graph capture #2075

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
valarLip merged 4 commits into from
Mar 3, 2026
Merged

Use unreg path for custom all-reduce during CUDA graph capture #2075

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
c19ff57
1
zyzshishui Feb 23, 2026
8146fca
1
zyzshishui Feb 24, 2026
4650e55
Merge branch 'main' into ar
zyzshishui Feb 24, 2026
a1d148f
1
zyzshishui Feb 25, 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
6 changes: 4 additions & 2 deletions aiter/dist/device_communicators/custom_all_reduce.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
* limitations under the License.
"""

import os
from contextlib import contextmanager
from typing import Any, List, Optional, Union

Expand Down Expand Up @@ -147,6 +148,7 @@ def __init__(
# return

self.disabled = False
self.tms_cudagraph = os.getenv("SGLANG_MEMORY_SAVER_CUDA_GRAPH", "0")

@valarLip valarLip Feb 23, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

prefix with SGLANG_MEMORY_SAVER_CUDA_GRAPH?

@zyzshishui zyzshishui Feb 23, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

Good catch. How about adding a tms_cudagraph parameter to __init__ and parse the param from sglang? But this would need sglang using updated aiter. Any suggestions?

@valarLip valarLip Feb 25, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

good idea, let's add "enable_register" param to init

@zyzshishui zyzshishui Feb 25, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

Added "enable_register_for_capturing" since this only control behavior for capturing not real calling. But again, we cannot make subsequent change before sglang bump up aiter's version. I will keep an eye on that

# buffers memory are owned by this Python class and passed to C++
# meta data composes of two parts: meta data for synchronization
# (256 bytes) and a temporary buffer for storing intermediate
Expand Down Expand Up @@ -329,8 +331,8 @@ def custom_all_reduce(
input,
use_new=use_new,
open_fp8_quant=open_fp8_quant,
registered_input=False,
registered_output=False
registered_input=not self.tms_cudagraph,
registered_output=not self.tms_cudagraph
)

def reduce_scatter(
Expand Down
Loading