bitxsw93 comments

Results 15 comments of


                                            bitxsw93

MNN android 平台 MNN_FORWARD_OPENCL使用gpu设备, inference 时间变长

你关注的是哪个模型？推理时间变长是指？

ubuntu cuda 输出全0

Please update lastest mnn code, retry again. If having any problem, please let me know.

MNN-2.0.0 release CUDA bacnend outputs error

Please update lastest mnn code, retry again. If having any problem, please let me know.

cuda后端编译成功,为啥推理时还不能用

Please update lastest mnn code, retry again. If having any problem, please let me know.

编译cuda出错

Please update lastest mnn code, retry again. If having any problem, please let me know.

UHDM模型cuda 显存比ORT大很多，耗时是ort的3倍

耗时主要是global avg pool导致的性能很慢。等内部修复同步吧。显存问题也会一并看下

MNN CUDA后端benchmark运行时间为零且无其他反馈

可以先采用CPU测试一下，保证下测试流程正确性。CUDA目前支持单输入的ConvolutionDepthwise

用OpenCL得到的输出不稳定

Please update lastest mnn code, retry again. If having any problem, please let me know.

Benchmark ?

> > Here's something interesting: I almost got a full generation on a 2gb 32bit mobile phone, before running out of ram. If someone has a better 32bit arm device,...

run diffusion demo error

用的是什么机器，内存有多大。可能是内存不够用了