Issues

• [Bug] Unauthenticated Mooncake STAGING_RSP chunk index causes Prefill scheduler OOM

  sgl-project/sglang#45014 · Filter by author yyymk opened on Oct 18, 2026

• [Bug] Unauthenticated Mooncake Decode registrations can exhaust Prefill memory and cause inference denial of service

  sgl-project/sglang#45012 · Filter by author yyymk opened on Oct 18, 2026

• [Feature] Allow DWDP with pipeline parallelism (--pp-size > 1)

  sgl-project/sglang#45011 · Filter by author QNan0513 opened on Oct 18, 2026

• [CI] test_deepseek_v41_flash_gb300_dp4ep4 fails on main: engram history int32 overflow from wide MM pad values (#43043)

  sgl-project/sglang#45010 · Filter by author nvpohanh opened on Oct 18, 2026

• [Bug] Client-supplied bootstrap address in Decode worker requests enables SSRF and blocks shared request processing

  sgl-project/sglang#45008 · Filter by author yyymk opened on Oct 18, 2026

• [Bug] Data-plane API key accepted on mesh HA management routes in sgl-model-gateway (control-plane authorization bypass)

  sgl-project/sglang#45006 · Filter by author yyymk opened on Oct 18, 2026
  1

• [Bug] Client-controlled RID prefix allows cross-request abort through disconnect cleanup

  sgl-project/sglang#45002 · Filter by author yyymk opened on Oct 18, 2026

• [Bug] FP8 block-quantized checkpoint on SM86 (FP8 Marlin) intermittently gives wrong tokens with the overlap scheduler; exact without it

  sgl-project/sglang#45000 · Filter by author DiaaAj opened on Oct 18, 2026
  1

• [Bug] Marlin MoE (AWQ W4A16) on SM86 intermittently returns wrong expert outputs, even for a single request; the Triton WNA16 kernel is exact

  sgl-project/sglang#44999 · Filter by author DiaaAj opened on Oct 18, 2026

• [Bug] NVFP4 Marlin fallback (SM80-SM90) garbles every output when the block scales are small (vllm-project/vllm#34577 not ported)

  sgl-project/sglang#44998 · Filter by author DiaaAj opened on Oct 18, 2026

• [Bug] NGRAM speculative decoding corrupts radix-cached KV with the overlap scheduler when requests end on a stop token (0.5.22)

  sgl-project/sglang#44997 · Filter by author DiaaAj opened on Oct 18, 2026

• [NPU] Please keep publishing Docker images for Ascend 910B

  sgl-project/sglang#44959 · Filter by author yuyicai opened on Oct 18, 2026

• DeepSeek V3.2/V4/V4.1 streaming tool-call parser: pre-existing bugs found while fuzzing

  sgl-project/sglang#44941 · Filter by author divyvasal opened on Oct 18, 2026

• [Bug] /v1/messages returns HTTP 500 for gpt-oss when the history contains a thinking block

  sgl-project/sglang#44932 · Filter by author yyming009 opened on Oct 17, 2026
  2

• [Bug] gpt-oss on /v1/chat/completions drops replayed reasoning: the template reads thinking, SGLang only forwards reasoning_content

  sgl-project/sglang#44931 · Filter by author yyming009 opened on Oct 17, 2026
  1

• [Bug] [AMD] --ep-dispatch-algorithm fake routes with the real correction bias on the aiter path

  sgl-project/sglang#44897 · Filter by author ch-wan opened on Oct 17, 2026
  3

• [Bug] Gemma4 input_embeds request crashes target prefill with input_ids/inputs_embeds exclusivity error

  sgl-project/sglang#44895 · Filter by author riversky0014 opened on Oct 17, 2026
  2

• [Feature] kv-shard: Pass Q instead of KV for short prefills with long cached prefixes

  sgl-project/sglang#44885 · Filter by author Shunkangz opened on Oct 17, 2026

• [Feature] Support pending external linker lookup at native prefill admission

  sgl-project/sglang#44872 · Filter by author feichai0017 opened on Oct 17, 2026

• [Feature] Prepare registered radix cache backends before worker graph initialization

  sgl-project/sglang#44855 · Filter by author feichai0017 opened on Oct 17, 2026

• [Bug] pause_generation(abort) + release_memory_occupation right after a client disconnect crashes scheduler

  sgl-project/sglang#44840 · Filter by author Yunzez opened on Oct 17, 2026
  1

• [Bug] With Speculative decoding + abort, completion can return more tokens than max_tokens

  sgl-project/sglang#44839 · Filter by author Yunzez opened on Oct 17, 2026

• [Bug] With pause_generation(in_place) then pause_generation(abort), the server hangs until killed by watchdog

  sgl-project/sglang#44835 · Filter by author Yunzez opened on Oct 17, 2026
  2

• [Bug] update_weights_from_disk after release_memory_occupation is accepted and crashes the scheduler (segfault in weight_loader)

  sgl-project/sglang#44834 · Filter by author Yunzez opened on Oct 17, 2026

• [CI] Post-merge runs on main cancel each other in NPU/XPU/Xeon/MUSA/SMG workflows

  sgl-project/sglang#44832 · Filter by author JerryChaox opened on Oct 17, 2026