Issues

• [Bug] Kunlunxin P800: Qwen3-0.6B produces incorrect output when FlagGems is enabled

  flagos-ai/vllm-plugin-FL#571 · Filter by author hooray03 opened on Oct 8, 2026
  

• [Bug] Kunlunxin P800: local Triton INT8 quantization test crashes with SIGSEGV

  flagos-ai/vllm-plugin-FL#570 · Filter by author hooray03 opened on Oct 8, 2026
  

• [Bug]: vllm_fl import fails on torch < 2.8 vendor builds that ship an unimportable torch.distributed._symmetric_memory

  bug
  flagos-ai/vllm-plugin-FL#550 · Filter by author tengqm opened on Sep 20, 2026

• [Bug]: [Iluvatar][vLLM 0.24] CUDAGraph capture_begin returns ixErrorNotSupported on BI-V150

  bug
  Bug

  flagos-ai/vllm-plugin-FL#539 · Filter by author hooray03 opened on Sep 17, 2026
  

• [Bug] Kunlunxin: Qwen3.6-27B graph-mode concurrent multimodal batch degenerates to repeated token 0

  bug
  flagos-ai/vllm-plugin-FL#529 · Filter by author tengqm opened on Sep 16, 2026

• 【FR】Bug: vllm-plugin-FL error: dispatch, vllm_fl on 沐曦(Metax) (Klear-46B-A2.5B-Instruct)

  bug
  flagos-ai/vllm-plugin-FL#524 · Filter by author Lxiparer opened on Sep 15, 2026

• [PERF]: optimize DeepSeek V4-W8A8

  enhancement
  Task

  flagos-ai/vllm-plugin-FL#523 · Filter by author ceci3 opened on Sep 15, 2026
  1

  

• 【FR】[Bug] KlearMoeForCausalLM fails to start under vllm-plugin-FL on 海光(Hygon): incomplete transformers-compat shim + hardcoded model registry (Kwai-Klear/Klear-46B-A2.5B-Instruct)

  flagos-ai/vllm-plugin-FL#522 · Filter by author ckxud opened on Sep 15, 2026
  1

• 【FR】Bug: 服务启动系统性挂起 — distributed init 之后卡死 (nvidia/AceInstruct-1.5B, 天数 Iluvatar)

  bugiluvatar
  flagos-ai/vllm-plugin-FL#521 · Filter by author Lxiparer opened on Sep 15, 2026

• 【FR】Bug: 生成阶段崩溃 — Mamba _C::selective_scan_fwd CUDA 内核缺失 (ai21labs/AI21-Jamba-Reasoning-3B, 天数 Iluvatar)

  bugiluvatar
  flagos-ai/vllm-plugin-FL#520 · Filter by author Lxiparer opened on Sep 15, 2026

• 【FR】Bug: vllm-plugin-FL error: vllm-plugin-FL on 华为(Ascend) (zai-org/GLM-4.7-Flash)

  ascendbugplugin-error
  flagos-ai/vllm-plugin-FL#519 · Filter by author Lxiparer opened on Sep 15, 2026

• 【FR】[Bug] vllm_fl ReferenceBackend missing moe_align_block_size → attention_backend has no fallback, service fails to start without full FlagGems op set on 海光(Hygon)

  flagos-ai/vllm-plugin-FL#517 · Filter by author ckxud opened on Sep 15, 2026

• 【FR】[Adaptation Gap] vllm-plugin-FL missing KlearMoeForCausalLM support — server crashes at startup (Kwai-Klear/Klear-46B-A2.5B-Instruct)

  flagos-ai/vllm-plugin-FL#515 · Filter by author jianchidundundun opened on Sep 15, 2026
  1

• [FR] Bug: vllm_fl plugin profile_run 与 Qwen3-VL PatchEmbed shape 不匹配导致 EngineCore 初始化崩溃 on 沐曦(Metax) (FINAL-Bench/Darwin-28B-Coder)

  P0
  flagos-ai/vllm-plugin-FL#471 · Filter by author Lxiparer opened on Sep 9, 2026

• [Ascend 910C] MiniCPM4-8B 服务崩溃：vllm-plugin-FL 显存探测 bug 导致 persistent OOM

  P0
  flagos-ai/vllm-plugin-FL#470 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: profile_run torch.compile 崩溃 non-function or method super: _disabled_torch_function_impl (Mthreads / Qwen3.5-27B-Derestricted)

  bugP0
  flagos-ai/vllm-plugin-FL#469 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: MoE 后端选择在 musa 平台下 UnboundLocalError: '_AVAILABLE_BACKENDS' referenced before assignment (Mthreads / inclusionAI/Ring-lite)

  bugP0
  flagos-ai/vllm-plugin-FL#468 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: Operator crash: mm/mm_out (mm_kernel) illegal memory access on 沐曦(Metax) (internlm/OREAL-32B)

  P0
  flagos-ai/vllm-plugin-FL#467 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: Operator crash (mm_out) on 沐曦(Metax) — vllm-plugin-FL (allenai/OLMo-2-0325-32B-Instruct)

  P0
  flagos-ai/vllm-plugin-FL#466 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: AscendMLABackend 未实现，不支持 MLA 架构模型 (moonshotai/Moonlight-16B-A3B-Instruct)

  华为(Ascend)bugP0plugin-error
  flagos-ai/vllm-plugin-FL#465 · Filter by author Lxiparer opened on Sep 9, 2026

• 【FR】Bug: MLA architecture FlashAttention backend missing on Iluvatar platform

  P0
  flagos-ai/vllm-plugin-FL#444 · Filter by author Lxiparer opened on Sep 4, 2026

• [FR] Bug: Service startup crash on 沐曦(Metax) – non-FlagGems, points to vllm-plugin-FL / framework (GLM-4.7-Flash)

  P0
  flagos-ai/vllm-plugin-FL#443 · Filter by author Lxiparer opened on Sep 4, 2026

• [Feature]: 缺少vllm中算子的FLagems实现

  enhancementP0
  flagos-ai/vllm-plugin-FL#438 · Filter by author tonyh168 opened on Sep 3, 2026

• [T-Head / PPU][MTP] Qwen3.8 FlashNext multi-hidden state is dropped by the vLLM 0.24 proposer path

  P0
  flagos-ai/vllm-plugin-FL#437 · Filter by author Cyrilw92 opened on Sep 3, 2026

• WorkerFL initialize_from_config breaks on vLLM 0.20.2: kv_cache_config list + missing set_current_vllm_config context

  P0
  flagos-ai/vllm-plugin-FL#419 · Filter by author tengqm opened on Aug 28, 2026