MNN
MNN copied to clipboard
alibaba
MNN is a blazing fast, lightweight deep learning framework, battle-tested by business-critical use cases in Alibaba. Full multimodal LLM Android App:[MNN-LLM-Android](./apps/Android/MnnLlmChat/README....
# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): Ubuntu 16.04 GCC-8.2 CUDA-10.2 MNN-Version: MNN-2.0.0 release # 编译方式: # Compiling Method -DCMAKE_BUILD_TYPE=Release -DMNN_CUDA=ON Model's output was wrong with MNN-2.0.0...
# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): 在sigmastar平台上编译 # Github版本: # Github Version: 1.2.6版本之后都编译失败，1.2.5之前正常 # 编译日志: # Build Log: /3rdparty/MNN/source/backend/cpu/compute/WinogradOptFunction.cpp:799:1: internal compiler error: in decompose_automod_address, at rtlanal.c:6311...
# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): aarch64-openwrt-linux-gun gcc 6.3.0 # Github版本: # Github Version: 1.2.10 在aarch64环境下，将两个模型导入到GPU上，会出现在第二个模型上崩溃，给出的错误日志 ``` [SignalHandler]catch signal: 11 [logGetBackTrace]string:linux-vdso.so.1(__kernel_rt_sigreturn+0) [0x7f78345770] [logGetBackTrace]string:/lib/libPVROCL.so(+0x39cc4) [0x7f7e390cc4] ``` 压测抓到过一帧日志是...
# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): Ubuntu linux 18.04 # Github版本: # Github Version: latest master till 2021.4.15 # 编译日志: # Build Log: 编译正常，可以用benchmark测试cuda和tensorrt。 ``` 从https://github.com/zeusees/License-Plate-Detector...
勾选了MNN_CUDA,MNN_USE_SYSTEM_LIB,取消了MNN_SEP_BUILD,ubuntu18上成功编译,推理时采用cuda后端还是报Can't Find type=2 backend, use 0 instead
work flow: 1. pip install -U mnn 2. generate fix input onnx from pytorch model: ``` torch.onnx.export(model, dummy_input, onnx_model_name, verbose=True, input_names=['input'], output_names=['output']) ``` 3. generate mnn module from onnx model:...
Hi, 非常感谢MNN的工作，很cool!!!!!!!!!! 我在windows平台编译了推理引擎，mnn版本是当前最新版，按照文档 https://www.yuque.com/mnn/cn/build_windows ``` cd /path/to/MNN powershell # 运行该命令从cmd环境进入powershell环境，后者功能更强大 ./schema/generate.ps1 # CPU+OpenCL+Vulkan, 64位编译 .\package_scripts\win\build_lib.ps1 -path MNN-CPU-OPENCL/lib/x64 -backends "opencl,vulkan" ``` 编译结果很流畅，只有几个warning: ```shell [65/316] Building CXX object CMakeFiles\MNNTransform.dir\source\geometry\GeometryTopK.cpp.obj xxx\MNN\source\geometry\GeometryTopK.cpp(19): warning C4477:...
hi 我在CommonOptFunction.cpp 中的函数中加上打印函数，但执行的时候无法打印出来，请问是什么原因呢？ （已经开了logcat，函数应该也调用了，但是log没打印） 谢谢
# 平台(如果交叉编译请再附上交叉编译目标平台): # Platform(Include target platform as well if cross-compiling): win10 # Github版本: # Github Version: 2.0.3 直接下载ZIP包请提供下载日期以及压缩包注释里的git版本(可通过``7z l zip包路径``命令并在输出信息中搜索``Comment`` 获得，形如``Comment = bc80b11110cd440aacdabbf59658d630527a7f2b``)。 git clone请提供 ``git commit`` 第一行的commit id Provide...
