Skip to content

Porting internode test from DeepEP to NixlEP - #1

Open
2dm wants to merge 2 commits into
mainfrom
internode_ep2
Open

Porting internode test from DeepEP to NixlEP#1
2dm wants to merge 2 commits into
mainfrom
internode_ep2

Conversation

@2dm

@2dm 2dm commented Jan 29, 2026

Copy link
Copy Markdown
Owner

Adding high throughput internode test to NIXL examples

Co-authored-by: Roey Azran <roeya@nvidia.com>
@@ -0,0 +1,23 @@
#pragma once
#ifdef ENABLE_DEBUG_LOGS

@2dm 2dm Jan 29, 2026

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

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

Consider removing logging completely. internode.cu is full of logging we previously added. The excess prints are not useful for debugging. It is also not aligned with elastic example and upstream DeepEp.


# Initialize NIXL buffer with group (for IPC handles) and TCPStore (for NIXL metadata)
print(f"pid: {os.getpid()}, rank: {rank}, num_ranks: {num_ranks}, initializing buffer", flush=True)
buffer = nixl_ep.Buffer(rank=rank, low_latency_mode=False, explicitly_destroy=True, group=group, tcp_store_group=tcp_store, nvlink_backend="ipc")

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

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

We have not supported low_latency_mode before. If we plan to support it, we should assign args.test_ll_compatibility to it.

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

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

Or consider to remove it completely from internode and diverge more from DeepEp

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant