Issues

• [Bug] Gemma4ForCausalLM loader ignores global_head_dim for full-attention layers — text-only Gemma‑4 A4B checkpoints fail to load (same weights load fine as Gemma4ForConditionalGeneration)

  sgl-project/sglang#43434 · Filter by author samidh opened on Oct 9, 2026

• [Feature] Add native SmolLM3 support

  sgl-project/sglang#43428 · Filter by author RachitSharma2001 opened on Oct 9, 2026

• Batch image outputs in server mode

  sgl-project/sglang#43418 · Filter by author OnePunchMonk opened on Oct 9, 2026
  1

• [Bug] --enable-deterministic-inference does not make hybrid Mamba2 models (granite-4.0-h) batch-invariant

  sgl-project/sglang#43413 · Filter by author Yunzez opened on Oct 9, 2026

• Speculative decoding crashes server on first non-greedy request: flashinfer verify-sampling op JIT fails (bundled cccl vs pip cu13 toolkit) and kills the process tree

  sgl-project/sglang#43412 · Filter by author santanu20 opened on Oct 9, 2026

• [RFC] Native worker batching for ComfyUI diffusion requests

  sgl-project/sglang#43411 · Filter by author OnePunchMonk opened on Oct 9, 2026

• [Bug] [diffusion] /v1/images/edits accepts a mask upload but silently ignores it

  sgl-project/sglang#43409 · Filter by author Shreyasg13 opened on Oct 9, 2026

• [Bug] CUDA VMM multimodal transport slice leaks when a request is aborted while still queued

  sgl-project/sglang#43402 · Filter by author StarDuster opened on Oct 9, 2026
  2

• ComfyUI-mode cannot load FLUX.1-schnell single-file checkpoints

  sgl-project/sglang#43395 · Filter by author OnePunchMonk opened on Oct 9, 2026

• [Feature] [Diffusion] Support LingBot-VLA 2.0

  sgl-project/sglang#43379 · Filter by author 0xtoward opened on Oct 9, 2026

• [Bug] runai_streamer: CPU-path tensors alias the streamer's staging buffers, so loaders that hold a tensor read overwritten bytes

  sgl-project/sglang#43352 · Filter by author fergusfinn opened on Oct 9, 2026

• [Feature] Support quality choices for Ascend NPU Diffusion

  sgl-project/sglang#43351 · Filter by author OrangeRedeng opened on Oct 9, 2026

• [Bug] RuntimeError: Failed at /sgl-workspace/sglang/python/sglang/kernels/jit/csrc/gemm/marlin/gptq_marlin_repack.cuh:311: size_n = 8608 is not divisible by tile_n_size = 64

  sgl-project/sglang#43342 · Filter by author frontword opened on Oct 9, 2026
  1

• [Bug] CuTe DSL GDN decode reads the state pool transposed, producing garbage output

  sgl-project/sglang#43331 · Filter by author SunskyXH opened on Oct 9, 2026

• [Bug] EP standard dispatch: -1 sentinel aliases to last local expert, padded spec-verify rows overflow masked-GEMM per-expert m_max (OOB stores, engine crash)

  sgl-project/sglang#43324 · Filter by author number-eleven11 opened on Oct 9, 2026

• MLLM Roadmap (2026 q4)

  Multi-modalroadmapvlm
  sgl-project/sglang#43314 · Filter by author mickqian opened on Oct 9, 2026

• SGLang-Diffusion Roadmap (2026 Q4)

  diffusionroadmap
  sgl-project/sglang#43313 · Filter by author mickqian opened on Oct 9, 2026
  1

  

• [Feature] Metrics to localize bottlenecks in the multi-tokenizer / multi-detokenizer IPC path

  sgl-project/sglang#43304 · Filter by author hope1262946533 opened on Oct 9, 2026

• [MegaMoE] W4A4 (mxf4xmxf4) target path for Kimi-K3: feature-completion patches + bs=1 perf/numerics findings

  sgl-project/sglang#43294 · Filter by author ZhimaoL opened on Oct 9, 2026
  1

• [Bug] EAGLE + DP attention: idle verify crashes with AttributeError: 'EagleVerifyInput' object has no attribute 'kv_indptr' (triton, eager)

  sgl-project/sglang#43287 · Filter by author StarDuster opened on Oct 9, 2026
  1

• [Bug] Glm47MoeDetector streaming emits invalid JSON for array/object/boolean values that aren't strict JSON ("1, 2", "True", empty)

  sgl-project/sglang#43273 · Filter by author Shreyasg13 opened on Oct 9, 2026
  1

• [Playground] Verified cell: h100 / default / fp8 / low-latency / single

  sgl-project/sglang#43263 · Filter by author skyegibson98-sketch opened on Oct 9, 2026

• [Performance] Cake NVFP4 warp-decode regresses TPOT on Qwen3-30B-A3B (GB200 TP1)

  sgl-project/sglang#43258 · Filter by author Smallfu666 opened on Oct 9, 2026

• [Feature] Support Aleph-Alpha/Kolibri-1

  sgl-project/sglang#43256 · Filter by author ischencheng opened on Oct 9, 2026

• [HiCache] NVFP4 KV + kernel io-backend: host pool row geometry mismatch and block scales never transported (packed MTP draft silently corrupted, accept length pinned to 1.0)

  sgl-project/sglang#43239 · Filter by author chengye75-star opened on Oct 8, 2026