Closed mahmoodn closed 3 years ago
They were moved to the deprecated branch and should be inside the darknet directory at level2
I cloned that branch and ran setup.sh. However, it seems that not enough. I also switched to src/cuda/level2/darknet and executed make and got this error
./src/convolutional_layer.c:167:13: error: ‘CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT’ undeclared (first use in this function)
167 | CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT,
Without knowing your execution environment, the answers here are merely speculative. This error is likely caused by CUDA 11 and cuDNN 8. It was tested with CUDA 9 and 10 with cuDNN 7 before and it's likely not supported by newer versions. There might be plans to address those issues but downgrading the CUDA and cuDNN might solve the problem.
Yes that was CUDA 11.2. Thanks.
Hi I don't see DNN workloads in level 0, 1 and 2 folders. I mean softmax, activation and ... Where should I look exactly?