Darknet: CUDNN_STATUS_BAD_PARAM

[koko-bap]

Hi, when i run ./darknet rnn train cfg/rnn.train.cfg -file /whatever.txt or ./darknet rnn generate cfg/rnn.cfg etc i get the error CUDNN_STATUS_BAD_PARAM for the line 234 in convolutional_layer.c:

CHECK_CUDNN(cudnnSetConvolutionGroupCount(l->convDesc, l->groups));

The error msg:

[sk@sk dark2]$ ./darknet rnn train cfg/rnn.train.cfg backup/rnn.backup -file /home/sk/dark2/new\ 2.txt 
rnn
   layer   filters  size/strd(dil)      input                output
   0 RNN Layer: 256 inputs, 512 outputs

 cuDNN status Error in: file: ./src/convolutional_layer.c : () : line: 234 : build time: Aug 14 2019 - 15:59:55 
cuDNN Error: CUDNN_STATUS_BAD_PARAM
cuDNN Error: CUDNN_STATUS_BAD_PARAM: File exists
darknet: ./src/utils.c:293: error: Assertion 0 failed.
Aborted (core dumped)

Ive tried with several older versions of CUDA (9.2 - 10.1) and CUDNN and also for both windows and linux, but nothing helped, except compiling without CUDNN.

Both training and detection works with CUDNN if using detector train/test/demo. Any help would be greatly appreciated!

(Ive been using a GTX1070 on the win10 box and RTX2060 on Fedora)

[326-T]

Hi, I have same problem. LSTM layer doesn't work.
Is there any progress?

[326-T]

I found a bug in connected_layer.c!
That is some function for convolutional_layer are loaded.
So, if you comment out these function while referring to the original code https://github.com/pjreddie/darknet, you can load the RNN layer.

But, LSTM layer still doesn't work!!
Is there another place to fix?
Thank you in advance!

[AlexeyAB]

@koko-bap @326-T I fixed it: https://github.com/AlexeyAB/darknet/commit/7bf83018e48ffaf3883f2020dedebb98929acf4f

[326-T]

@AlexeyAB
Thank you!!!
It works.