Repository Issue Activity (beta)

#344 NVL/RDMA tuning process (auto tune)

#670 [BUG] DeepEPv2 internode num_worst_tokens tests

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

#721 Observation of Emergent Behavior in DeepSeek Model (Anomaly Type‑7)

#733 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)

#227 Issue 1: Missing docstring for internode_combine method

#717 误删长期重要对话，邮件发送至[email protected]一周未获回复，恳请协助

#726 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)

#725 Travel documents and visa proc

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

#13 Test fail use 2node on tests/test_internode.py

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

#700 Low-latency combine returns one corrupted token row on B200 (~3% of invocations at max tokens/rank)

#710 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

#567 question about Hybrid-EP

#698 `hybrid_ep_cpp` extension hardcodes `-std=c++17`; should inherit torch's C++ standard like `deep_ep_cpp`

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

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

#689 [Proposal] deterministic hybrid dispatch implementation

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

#691 [Proposal] Adding HybridEP compatibility with newer NVSHMEM versions

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

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

#628 ElasticBuffer asserts NCCL_GIN_TYPE_NONE on host with MLNX_OFED 24.10 (libmlx5 < MLX5_1.25)

#624 [Performance] Warn if automatic NUMA balancing of the Linux kernel is enabled