Issues

• [Feature] Per-node HTTP frontend for data-parallel groups spanning nodes, addressed by global X-Data-Parallel-Rank

  sgl-project/sglang#47657 · Filter by author robertgshaw2-redhat opened on Oct 31, 2026

• [RFC] NPU RL Test Consolidation: File Ownership, Coverage Docstrings, and CI Integration (Ascend)

  sgl-project/sglang#47624 · Filter by author SeaAsters opened on Oct 31, 2026

• [Bug] PP disaggregated decode produces garbage (runs of token 0) when pp_async_batch_depth > pp_size / 2

  sgl-project/sglang#47613 · Filter by author Phoenix3334 opened on Oct 31, 2026
  1

• [Test] Cover model-loader extra config validation

  sgl-project/sglang#47607 · Filter by author pratik-naik003 opened on Oct 31, 2026

• [Test] Cover unindexed Mistral safetensors filtering

  sgl-project/sglang#47602 · Filter by author pratik-naik003 opened on Oct 31, 2026

• [Bug] rust-server: attention-DP listeners on one CPU mask pin their frontends to the same cores

  sgl-project/sglang#47598 · Filter by author phphlio1 opened on Oct 31, 2026

• [Test] Add CPU unit coverage for MTP weight filtering

  sgl-project/sglang#47594 · Filter by author pratik-naik003 opened on Oct 31, 2026

• [AITER-Upgrade] PR readiness gfx1250 (MI45x)

  sgl-project/sglang#47562 · Filter by author kangwangamd opened on Oct 31, 2026

• [Bug] rust-server: /internal/peer serves a relayed control request without checking it was meant for this DP rank

  sgl-project/sglang#47511 · Filter by author metamergebot opened on Oct 30, 2026

• [Bug] rust-server: gather_server_endpoints can advertise a loopback or unroutable address to other nodes, and only warns

  sgl-project/sglang#47509 · Filter by author metamergebot opened on Oct 30, 2026

• [Bug][ROCm] Kimi-K3 fused KDA input-projection buffers are replaced on weight reload

  sgl-project/sglang#47508 · Filter by author wolfcomos opened on Oct 30, 2026

• [Bug] rust-server: gather_server_endpoints runs on elastic-EP joiners over a WORLD group the serving ranks never enter

  sgl-project/sglang#47505 · Filter by author shodoco opened on Oct 30, 2026

• [Bug] Qwen3.8-Flash-Next-FP8 TP4/EP4 decode ~5% slower on GB200 since #45917 (FP8 TRT-LLM MoE leaves in-kernel routing under EP)

  sgl-project/sglang#47501 · Filter by author nvpohanh opened on Oct 30, 2026

• [Bug] EAGLE draft extend crashes with negative dimension under DP attention + FlashInfer A2A + prefill CUDA graph

  sgl-project/sglang#47490 · Filter by author mattteochen opened on Oct 30, 2026

• [Performance] Reproducible cache hit-rate dip during L2/L3 capacity pressure with HiCache + Mooncake

  sgl-project/sglang#47481 · Filter by author banfeb opened on Oct 30, 2026

• stop_regex: a non-literal match is left in the returned text, and no_stop_trim has no effect for regex stops

  sgl-project/sglang#47461 · Filter by author yzhang90 opened on Oct 30, 2026

• [Bug] NemotronHConfig rejects transformers 5 layer types linear_attention / full_attention in config.json

  sgl-project/sglang#47444 · Filter by author janbernloehr opened on Oct 30, 2026

• [Bug] S3 model URI overrides explicit remote_instance + modelexpress with runai_streamer

  sgl-project/sglang#47428 · Filter by author zhy76 opened on Oct 30, 2026

• [Bug] benchmark: --dataset-name random silently requires the ShareGPT download; fails on offline/restricted networks

  sgl-project/sglang#47420 · Filter by author DonOffice opened on Oct 30, 2026

• [Bug] Attention-DP ranks are indistinguishable to routers and dashboards: KvMetrics.data_parallel_rank is always 0 and Prometheus series omit dp_rank

  sgl-project/sglang#47409 · Filter by author Swipe4057 opened on Oct 30, 2026

• [Bug] [NPU] 图模式下走Fia v2算子，torch_npu.npu_fused_infer_attention_score_v2函数入参错误

  sgl-project/sglang#47399 · Filter by author luobicangqiong opened on Oct 30, 2026

• SGLANG_JIT_DEEPGEMM_FAST_WARMUP on DeepSeek-V4.1: 3-5x faster cold start, no runtime degradation detectable at chunked-prefill 16384

  sgl-project/sglang#47397 · Filter by author xulunfan opened on Oct 30, 2026
  2

• [Doc][NPU] Various missing clarifications, prerequisites and validation steps in Ascend NPU documentation

  sgl-project/sglang#47382 · Filter by author axx-ty911 opened on Oct 30, 2026

• Llama KV scale loader replaces tensor scales with floats and leaves host mirrors stale

  sgl-project/sglang#47357 · Filter by author EdwardZhang1108 opened on Oct 30, 2026

• [Bug] diffusion: LingBot-Video-MoE crashes in text encoding under default layerwise offload (tied vocab table on CPU)

  sgl-project/sglang#47297 · Filter by author OnePunchMonk opened on Oct 29, 2026