[GH-ISSUE #271] A Small Question Regarding Test Script `test_low_latency.py` #201

New issue

Closed

opened 2026-10-26 09:56:27 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 09:56:27 +00:00

Owner

Originally created by @ZhiyiHu1999 on GitHub (Jul 1, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/271

Hello! I think in the hook case of low latency mode, the 'hook()' happens on the compute steam, so later computation enqueued on the same stream can read the correct data.

In the test_low_latency.py, it checks the received data on the CPU side as long as the hook() returns. However, to my comprehension, when the hook() returns, it only means that the LL recv kernel is launched. So, how is it ensured to read the correct data on the CPU side after the hook() returns instantly.

github.com/deepseek-ai/DeepEP@8b0c59448b/tests/test_low_latency.py (L93-L98)

Thanks a lot!

Originally created by @ZhiyiHu1999 on GitHub (Jul 1, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/271 Hello! I think in the hook case of low latency mode, the 'hook()' happens on the compute steam, so later computation enqueued on the same stream can read the correct data. In the `test_low_latency.py`, it checks the received data on the CPU side as long as the `hook()` returns. However, to my comprehension, when the `hook()` returns, it only means that the LL recv kernel is launched. So, how is it ensured to read the correct data on the CPU side after the `hook()` returns instantly. https://github.com/deepseek-ai/DeepEP/blob/8b0c59448bd808366d16fbcf036a231de9c5d0d2/tests/test_low_latency.py#L93-L98 Thanks a lot!

bsakaguchi closed this issue

2026-10-26 09:56:27 +00:00

bsakaguchi referenced this issue

2026-10-26 10:01:02 +00:00

[PR #201] [MERGED] remove the dependency of gdrcopy #463