elevenxiang comments

Results 5 comments of


                                            elevenxiang

Question about ib_write_lat with CUDA

> I've found that `ib_write_lat` doesn't support CUDA mode. Wonder whether there is any intrinsic issue that prevents supporting this? I think it should not be CUDA issue because NCCL...

[intranode] when num_tokens change to 65536, then check failed

skip recv check, and also found topk_weight check failed assert calc_diff(check_topk_weights, ref_topk_weights) < 1e-9 AssertionError

[intranode] when num_tokens change to 65536, then check failed

nobody ever changed the num_tokens to 64K ?

When running the two-machine test_low_latency.py (EP16), there is a significant difference in the test results between two machine

> This seems very similar to [#183](https://github.com/deepseek-ai/DeepEP/issues/183). We haven’t encountered this problem, but my guess is that it might be caused by misaligned kernel launch times. You can try generating...

When running the two-machine test_low_latency.py (EP16), there is a significant difference in the test results between two machine

> > I discussed this issue with [@liuhe-spec](https://github.com/liuhe-spec) on WeChat, and we strongly suspect it is likely related to RoCE network congestion control. > > If possible, you can ask...