Issues

• Compatibility issues with NVSHMEM version 3.5

  deepseek-ai/DeepEP#764 · Filter by author gzshan opened on Sep 23, 2026

• [Performance] DeepEP V2 combine epilogue low batch performance

  deepseek-ai/DeepEP#760 · Filter by author elvircrn opened on Sep 22, 2026

• [Bug] low-latency (internode_ll) unusable on non-Mellanox NICs: NVSHMEM IBGDA transport is mlx5-only (reproduced on Alibaba eRDMA)

  deepseek-ai/DeepEP#759 · Filter by author luoyuctl opened on Sep 21, 2026

• get_env<int> returns an uninitialized value when an env var is set to a non-numeric string (EP_BUFFER_DEBUG=false reads as true)

  deepseek-ai/DeepEP#733 · Filter by author tonydzi opened on Aug 21, 2026

• get_nccl_comm_handle reuses a null _comm_ptr() when the group's NCCL comm is lazily uninitialized — ncclTeamWorld null-deref at ElasticBuffer construction (vLLM serve, 16/16 ranks)

  deepseek-ai/DeepEP#726 · Filter by author dmvevents opened on Aug 14, 2026

• Legacy normal kernels trap for hidden >= 8192: TMA staging buffer caps hidden with no host-side check

  deepseek-ai/DeepEP#722 · Filter by author yashkgp opened on Aug 12, 2026

• EP_BUFFER_DEBUG=0 enables Python-side debug output because the env var is tested as a raw string

  deepseek-ai/DeepEP#718 · Filter by author yashkgp opened on Aug 9, 2026
  1

• pip-installing vLLM/torch silently downgrades NCCL below the 2.30.4 floor; import dies at undefined symbol and check_nccl_so() cannot catch it

  deepseek-ai/DeepEP#710 · Filter by author dmvevents opened on Jul 30, 2026

• DeepEP v1 Normal mode, run test_internode RuntimeError: DeepEP error: timeout (dispatch CPU) when use NVSHMEM version >= 3.5.19

  deepseek-ai/DeepEP#707 · Filter by author alpha-baby opened on Jul 28, 2026

• [Question] CP is in the README but isn't published

  deepseek-ai/DeepEP#694 · Filter by author KareemMusleh opened on Jul 22, 2026
  1

• [Proposal] Adding HybridEP compatibility with newer NVSHMEM versions

  deepseek-ai/DeepEP#691 · Filter by author youngjeong46 opened on Jul 15, 2026
  1

• [Proposal] deterministic hybrid dispatch implementation

  deepseek-ai/DeepEP#689 · Filter by author fishautumn opened on Jul 14, 2026
  3

• JIT cache: rename-race loser crashes with EP_HOST_ASSERT(runtime != nullptr) when the cache is on a shared filesystem (NFS)

  deepseek-ai/DeepEP#683 · Filter by author Oseltamivir opened on Jul 11, 2026

• [issue] st.bulk instruction requires sm_100, don't support pre-sm100 arch

  deepseek-ai/DeepEP#675 · Filter by author huakainomore opened on Jul 3, 2026

• [Performance] Question about performance of Deepep v2, the role of the prefer-overlap-with-compute parameter

  deepseek-ai/DeepEP#669 · Filter by author shenhongqian opened on Jun 22, 2026
  3

• Question: Why did DeepEP v2 drop the double-buffer design in favor of explicit pre/post-sync barriers?

  deepseek-ai/DeepEP#668 · Filter by author mkbk-with-circle opened on Jun 19, 2026
  1

• What is the function of check_fast_rdma_atomic_support

  deepseek-ai/DeepEP#667 · Filter by author shenhongqian opened on Jun 16, 2026

• Would deepep work in a hybrid GPU networking topology inside a single node(intronode)?

  deepseek-ai/DeepEP#666 · Filter by author shrutimohanty opened on Jun 15, 2026

• [Question] What's policy about apply ElasticBuffer for prefill?

  deepseek-ai/DeepEP#662 · Filter by author xiaoqi35 opened on Jun 11, 2026

• Can HybridEP NIXL path support EFA/libfabric, or is it UCX/GDA-specific?

  deepseek-ai/DeepEP#659 · Filter by author irenedea opened on Jun 3, 2026
  2

• Does NVSHMEM/DeepEP support SR-IOV Virtual Function devices?

  deepseek-ai/DeepEP#658 · Filter by author hmthhh opened on Jun 3, 2026
  2

• [DeepEPv2] test_agrs.py failed due to CUDA_ERROR_NOT_SUPPORTED, operation not supported

  deepseek-ai/DeepEP#656 · Filter by author elevenxiang opened on Jun 1, 2026
  1

• Zero-copy intranode for EPv2

  deepseek-ai/DeepEP#655 · Filter by author usernamehaha2022 opened on Jun 1, 2026

• [Performance] Question About Dispatch Prologue Barrier and Excessive P99 ITL With V2

  deepseek-ai/DeepEP#652 · Filter by author elvircrn opened on May 30, 2026

• [Question] Why is elastic dispatch limited to ~512 warps?

  deepseek-ai/DeepEP#636 · Filter by author lixingyii opened on May 19, 2026