Please provide the following information when requesting support.
• Hardware: x86 integrated Intel GPU
• Network Type: Detectnet_v2
• docker_tag: 4.0.0-pyt
Hello,
i am trying to convert the PeopleNet model (deployable_quantized_v2.6.1) with the tao-converter from tao-toolkit:4.0.0-pyt to use it with the Triton Inferenece Server.
First i tried
/opt/nvidia/tools/tao-converter resnet34_peoplenet_int8.etlt -k tlt_encode -p input_1,1x3x544x960,4x3x544x960,16x3x544x960 -e model.plan -t fp16 which results in the following error:
Error: no input dimensions given
but does work with adjusted paramateres for models like GestureNet, ActionRecognitionNet2D and others.
Then i tried:
/opt/nvidia/tools/tao-converter resnet34_peoplenet_int8.etlt -k tlt_encode -p input_1,1x3x544x960,4x3x544x960,16x3x544x960 -e model.plan -t fp16 -d 1x3x544x960,4x3x544x960,16x3x544x960
which also results in the following error:
[INFO] [MemUsageChange] Init CUDA: CPU +202, GPU +0, now: CPU 214, GPU 458 (MiB) [INFO] [MemUsageChange] Init builder kernel library: CPU +121, GPU +22, now: CPU 387, GPU 480 (MiB) [WARNING] CUDA lazy loading is not enabled. Enabling it can significantly reduce device memory usage. See CUDA_MODULE_LOADING in https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#env-vars [WARNING] The implicit batch dimension mode has been deprecated. Please create the network with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag whenever possible. [ERROR] 3: conv1/convolution:kernel weights has count 9408 but 3136 was expected [ERROR] 4: conv1/convolution: count of 9408 weights in kernel, but kernel dimensions (7,7) with 1 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 1 * 7*7 * 64 / 1 = 3136 [ERROR] 4: [convolutionNode.cpp::computeOutputExtents::58] Error Code 4: Internal Error (conv1/convolution: number of kernel weights does not match tensor dimensions) [ERROR] 3: conv1/convolution:kernel weights has count 9408 but 3136 was expected [ERROR] 4: conv1/convolution: count of 9408 weights in kernel, but kernel dimensions (7,7) with 1 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 1 * 7*7 * 64 / 1 = 3136 [ERROR] 4: [convolutionNode.cpp::computeOutputExtents::58] Error Code 4: Internal Error (conv1/convolution: number of kernel weights does not match tensor dimensions) [ERROR] 3: conv1/convolution:kernel weights has count 9408 but 3136 was expected [ERROR] 4: conv1/convolution: count of 9408 weights in kernel, but kernel dimensions (7,7) with 1 input channels, 64 output channels and 1 groups were specified. Expected Weights count is 1 * 7*7 * 64 / 1 = 3136 [ERROR] 4: [convolutionNode.cpp::computeOutputExtents::58] Error Code 4: Internal Error (conv1/convolution: number of kernel weights does not match tensor dimensions) [ERROR] UffParser: Parser error: conv1/BiasAdd: The input to the Scale Layer is required to have a minimum of 3 dimensions. [ERROR] Failed to parse the model, please check the encoding key to make sure it's correct [ERROR] 4: [network.cpp::validate::2738] Error Code 4: Internal Error (Network must have at least one output) [ERROR] Unable to create engine Segmentation fault
How do i convert the PeopleNet model to use it in Triton Inferenece Server with tensorrt as backend?
Thanks!