search

tusen-ai / simpledet

A Simple and Versatile Framework for Object Detection and Instance Recognition
Apache License 2.0
3.08k stars 486 forks source link

There is no method "mx.sym.contrib.BroadcastScale" in mxnet #350

Closed rwecho closed 3 years ago

rwecho commented 3 years ago

Hi, I bought Nvidia 3070 to run simpledet. But mxnet can not run with cuda 10.1 on 3070. To compile mxnet, I try to find the source corresponding to simpledet. I can not found the gen_contrib.py in the history of mxnet 1.6. So is there anyone working with the new Graphics card? Thanks.

rwecho commented 3 years ago

compare apache-mxnet-package with simpledet-package

There are some file not exists in apache

I'm new here, did not how gen_xx.py inject to mxnet package when compiling for cuda 11. Help, please.

huangzehao commented 3 years ago

Similar to https://github.com/TuSimple/simpledet/issues/351. Is your cuda version 11.1? We will build an image for it.

rwecho commented 3 years ago

yes, I'm planning to run on cuda 11.1. thanks.

huangzehao commented 3 years ago

@rwecho We provide a dockerfile in https://github.com/TuSimple/simpledet/tree/master/docker. I test the docker on gpu 3090 with cuda11.1 and cudnn8. It works.

rwecho commented 3 years ago

@huangzehao thanks, I'll test on my Nvidia 3070, and commit the result here.

rwecho commented 3 years ago

@huangzehao I test the dockerfile and failed. it warning as below:

nvcc fatal   : Unsupported gpu architecture 'compute_80'
nvcc fatal   : Unsupported gpu architecture 'compute_86'
huangzehao commented 3 years ago

I guess your cuda version is 11.0 instead of 11.1? You can try update cuda from 11.0 to 11.1 or remove -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 from echo 'CUDA_ARCH = -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86' >> ./config.mk && \ in dockerfile

rwecho commented 3 years ago

The Cuda version in the dockerfile specified is 10.2, I built the docker image with GitHub action. How did you build with the dockerfile ?

huangzehao commented 3 years ago

https://github.com/TuSimple/simpledet/tree/master/docker#build

rwecho commented 3 years ago

https://github.com/TuSimple/simpledet/tree/master/docker#build

Ignore the build args, I'll try it again. thanks.