MNN · Issues· 35 open
Open on GitHubLocally synced open issues (discussions stay on GitHub)
- #4890
[Bug] Omni reuses stale PLE input across successive text requests
Updated Sep 18, 2026 - #4887
[Feature/Optimization] OvisOCR2 (finetuned Qwen3.5-0.8B) vision encoder inference latency is significantly high on MNN
Updated Sep 18, 2026 - #4886
[Bug] Omni inherits embedding attention masks, breaking Gemma 4 mixed-attention generation
Updated Sep 18, 2026 - #4883
qwen3.5的mnn模型如何进行gpu推理
Updated Sep 17, 2026 - #4811
Android API inference crash: SIGABRT in MNN::DeferBufferAllocator::apply() / MNN::Session::resize()
type:bugplatform:androidmodule:llmUpdated Sep 17, 2026 - #4849
MNN 模型使用getSessionInfo 获取的内存数据和实测dumpsys meminfo相差较大
type:questionplatform:androidUpdated Sep 17, 2026 - #4802
Qwen2.5-Omni-3B/7B 在最新代码上opencl上推理崩溃
type:bugbackend:openclmodule:llmUpdated Sep 15, 2026 - #4859
[Feature Request][Android] MNN安卓应用更新并集成Hexagon NPU
Updated Sep 11, 2026 - #4815
Android 内存泄露
type:questionplatform:androidUpdated Sep 11, 2026 - #4779
[Model Request] Ling 3.0 Tiny
module:llmtype:model-requestUpdated Sep 11, 2026 - #4833
[Android] Fatal crash on SanaSession.initNative - MNN Chat 0.8.3 (P0)
type:bugplatform:androidapp:transformerUpdated Sep 11, 2026 - #4778
[Feature Request]: Gemma 4 12B
module:llmtype:model-requestUpdated Sep 11, 2026 - #4624
[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
type:bugplatform:androidbackend:openclmodule:llmapp:transformerUpdated Sep 11, 2026 - #4792
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
type:bugplatform:androidmodule:llmUpdated Sep 11, 2026 - #4716
[QNN] 使用 QNN 后端转换 Qwen3-VL-2B 时，`compilefornpu` 发生段错误。
type:bugplatform:androidmodule:llmbackend:qnnUpdated Sep 11, 2026