Search results

Hybrid_EP Abnormal performance on B300 (2-node dispatch ~40GB/s) reveals hardcoded DEF_IB_TC; request env-based configurability (align with NVSHMEM)

Status: Open.

#608 In deepseek-ai/DeepEP;

· kzlxd opened on Apr 24, 2026

Hybrid-EP has hang and precision issues in SFT training with variable sequence lengths

Status: Open.

#604 In deepseek-ai/DeepEP;

· fengchen-98 opened on Apr 17, 2026

[Feature Request] Support Expert-Choice Routing via Native Expert-Centric API ([num_experts, capacity])

Status: Open.

#595 In deepseek-ai/DeepEP;

· niyunsheng opened on Mar 30, 2026

HIddensize alignment can be different in e8m0/non e8m0 mode?

Status: Open.

#593 In deepseek-ai/DeepEP;

· MARD1NO opened on Mar 18, 2026

Poor HybridEP performance on dual-node 16x B300 GPUs - IB bandwidth significantly lower than expected

Status: Open.

#590 In deepseek-ai/DeepEP;

· sssssakuraaaaa opened on Mar 12, 2026

buffer.combine() produces NaN when torch.use_deterministic_algorithms(True) due to stream race on output tensor

Status: Open.

#589 In deepseek-ai/DeepEP;

· wyc-ruiker opened on Mar 8, 2026

Why doesn't low-latency dispatch use system-level fence before issuing IBGDA writes?

Status: Open.

#587 In deepseek-ai/DeepEP;

· hexclover opened on Mar 2, 2026

deep_ep.cpp:200 init fails on partial CUDA peer access; preflight and report unsupported device pairs (cudaErrorPeerAccessUnsupported)

Status: Open.

#584 In deepseek-ai/DeepEP;

· stewtong opened on Feb 20, 2026

AssertionError: No NVLink connection between GPU 0 and GPU 0

Status: Open.

#582 In deepseek-ai/DeepEP;

· pramodk opened on Feb 17, 2026

DeepEP not work on GB200 NVL72

Status: Open.

#579 In deepseek-ai/DeepEP;

· wjx985 opened on Feb 4, 2026

Question about synchronization guarantee in DeepEP dispatch low latency kernel

Status: Open.

#577 In deepseek-ai/DeepEP;

· Yi-sir opened on Feb 2, 2026

[Performance] B200 underperforms H100 in DeepEP Low-Latency Mode

Status: Open.

#576 In deepseek-ai/DeepEP;

· kzlxd opened on Jan 30, 2026