Issues

• CPU decode collapsed on Snapdragon 8 Elite Gen 5 (2x S26 Ultra, 1.6 tok/s) while prefill/OpenCL are healthy; same build does 8.9 tok/s on S25 Ultra

  alibaba/MNN#4922 · Filter by author Mr-J-369 opened on Oct 1, 2026
  1

• 邀请 mnn 加入 GithubStarMate，让更多人发现你的作品

  alibaba/MNN#4921 · Filter by author walrus3547 opened on Sep 30, 2026

• Error during generation in the Sana-Edit model.

  alibaba/MNN#4920 · Filter by author alexdsh opened on Sep 30, 2026

• share unwrapping cl_mem of Interpreter for Zero-Copy

  backend:opencltype:feature
  alibaba/MNN#4915 · Filter by author fatalfeel opened on Sep 29, 2026
  

• crash

  app:transformerplatform:androidstatus:needs-infotype:bug
  alibaba/MNN#4908 · Filter by author zeskadalton-hbz opened on Sep 23, 2026
  

• 创建推理会话时模型张量名进入默认哈希表的潜在哈希碰撞 CPU 放大

  status:needs-infotype:performance
  alibaba/MNN#4901 · Filter by author K-ANOY opened on Sep 22, 2026
  

• [Bug] Eagle 参数导出使用硬编码

  module:llmtype:bug
  alibaba/MNN#4895 · Filter by author huluhuluu opened on Sep 20, 2026
  

• [Feature/Optimization] OvisOCR2 (finetuned Qwen3.5-0.8B) vision encoder inference latency is significantly high on MNN

  module:llmstatus:needs-infotype:performance
  alibaba/MNN#4887 · Filter by author dr3334 opened on Sep 18, 2026
  

• qwen3.5的mnn模型如何进行gpu推理

  backend:cudamodule:llmmodule:pymnnplatform:windowstype:question
  alibaba/MNN#4883 · Filter by author 1457209720 opened on Sep 17, 2026
  8

  

• [Feature Request][Android] MNN安卓应用更新并集成Hexagon NPU

  app:transformerbackend:qnnmodule:llmplatform:androidtype:feature
  alibaba/MNN#4859 · Filter by author ArkaneFans opened on Sep 11, 2026
  1

  

• [Bug][Windows] gemma-4-E4B-it-MNN: CPU divide-by-zero crash and Vulkan segfault on Windows x64

  backend:vulkanmodule:llmplatform:windowstype:bug
  alibaba/MNN#4835 · Filter by author kangdy05 opened on Sep 8, 2026
  

• [Android] Fatal crash on SanaSession.initNative - MNN Chat 0.8.3 (P0)

  app:transformerplatform:androidtype:bug
  alibaba/MNN#4833 · Filter by author kongkong158205519 opened on Sep 5, 2026
  1

  

• Android 内存泄露

  platform:androidtype:question
  alibaba/MNN#4815 · Filter by author weixiao888 opened on Aug 27, 2026
  4

  

• Android API inference crash: SIGABRT in MNN::DeferBufferAllocator::apply() / MNN::Session::resize()

  module:llmplatform:androidtype:bug
  alibaba/MNN#4811 · Filter by author petaz1987 opened on Aug 26, 2026
  3

  

• Qwen2.5-Omni-3B/7B 在最新代码上opencl上推理崩溃

  backend:openclmodule:llmtype:bug
  alibaba/MNN#4802 · Filter by author Rye426 opened on Aug 24, 2026
  13

  

• 希望 MNN 支持 MTP（Multi-Token Prediction）模型的导出和推理适配

  module:llmtype:feature
  Feature

  alibaba/MNN#4793 · Filter by author Rye426 opened on Aug 21, 2026
  

• 已获得支持mtk npu 的最新仓库mnn3.6.0，使用neuropilot-sdk-basic-7.0.8-build20240807进行天玑9300/8300 llmexport.py导出后，使用neruonsdk 对第一层prompt编译报错

  module:llmplatform:androidtype:bug
  alibaba/MNN#4792 · Filter by author dr3334 opened on Aug 20, 2026
  2

  

• [Model Request] Ling 3.0 Tiny

  module:llmtype:model-request
  alibaba/MNN#4779 · Filter by author ThatHungarian opened on Aug 16, 2026
  1

  

• [Feature Request]: Gemma 4 12B

  module:llmtype:model-request
  alibaba/MNN#4778 · Filter by author enderdragonenforcer-source opened on Aug 16, 2026
  1

  

• [QNN] 使用 QNN 后端转换 Qwen3-VL-2B 时，compilefornpu 发生段错误。

  backend:qnnmodule:llmplatform:androidtype:bug
  alibaba/MNN#4716 · Filter by author yydhYYDH opened on Aug 5, 2026
  7

  

• ［bug］Gemma-4-E2B-it-MNN在识别图片时崩溃

  app:transformermodule:llmplatform:androidtype:bug
  alibaba/MNN#4641 · Filter by author harryz7654 opened on Jul 18, 2026
  6

  

• [bug]OpenCL后端在我的手机上运行，AI大模型时会出现溢出和参数不匹配

  app:transformerbackend:openclmodule:llmplatform:androidtype:bug
  alibaba/MNN#4624 · Filter by author harryz7654 opened on Jul 14, 2026
  4

  

• 英特尔gpu 采用opencl 加速异常

  backend:openclplatform:windowstype:bug
  alibaba/MNN#4565 · Filter by author NikoLXQ opened on Jun 24, 2026
  4