failed to inference about torch2trt HOT 3 CLOSED

i'm sorry, but,did you find the reason? i faced a similar issue:
[E] [TRT] ../rtSafe/cuda/cudaPoolingRunner.cpp (200) - Cudnn Error in execute: 3 (CUDNN_STATUS_BAD_PARAM)

from torch2trt.

i'm sorry, but,did you find the reason? i faced a similar issue: [E] [TRT] ../rtSafe/cuda/cudaPoolingRunner.cpp (200) - Cudnn Error in execute: 3 (CUDNN_STATUS_BAD_PARAM)

I got the same error, have you solved it? Could you give me some advice?Thanks.

from torch2trt.

i'm sorry, but,did you find the reason? i faced a similar issue: [E] [TRT] ../rtSafe/cuda/cudaPoolingRunner.cpp (200) - Cudnn Error in execute: 3 (CUDNN_STATUS_BAD_PARAM)

I got the same error, have you solved it? Could you give me some advice?Thanks.

Yeah, I'll show u my step(I test a craft.onnx in my case):
Trying process:
./trtexec --verbose --onnx=/tmp/trt/craft_modify_fix.onnx
Success

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx
[E] [TRT] Network has dynamic or shape inputs, but no optimization profile has been defined.

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx --explicitBatch
[E] [TRT] ../rtSafe/cuda/cudaPoolingRunner.cpp (200) - Cudnn Error in execute: 3 (CUDNN_STATUS_BAD_PARAM)
then u should refer:
https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#work_dynamic_shapes
https://github.com/NVIDIA/TensorRT/tree/f5e8b8c55060447c4d3a45e7379bc25374edc93f/samples/opensource/trtexec
to learn something

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx --explicitBatch --device=1 --dumpOutput --dumpProfile --useCudaGraph --workspace=2048 --minShapes=input:1x3x200x200 --optShapes=input:1x3x500x500 --maxShapes=input:1x3x800x800 --shapes=input:1x3x500x500
[E] [TRT] ../rtSafe/cuda/cudaConvolutionRunner.cpp (362) - Cudnn Error in execute: 9 (CUDNN_STATUS_NOT_SUPPORTED)

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx --explicitBatch --device=1 --dumpOutput --dumpProfile --useCudaGraph --workspace=2048 --minShapes=input:1x3x200x200 --optShapes=input:1x3x608x608 --maxShapes=input:1x3x800x800 --shapes=input:1x3x500x500
[E] [TRT] Concat_150: dimensions not compatible for concatenation

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx --explicitBatch --device=1 --dumpOutput --dumpProfile --useCudaGraph --workspace=2048 --minShapes=input:1x3x200x200 --optShapes=input:1x3x608x608 --maxShapes=input:1x3x800x800 --shapes=input:1x3x608x608
[E] [TRT] ../rtSafe/safeContext.cpp (133) - Cudnn Error in configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

./trtexec --verbose --onnx=/tmp/trt/craft_modify_dynamic.onnx --explicitBatch --dumpOutput --dumpProfile --useCudaGraph --workspace=2048 --minShapes=input:1x3x200x200 --optShapes=input:1x3x608x608 --maxShapes=input:1x3x800x800 --shapes=input:1x3x608x608
Success

In conclusion, u have to learn the useage from https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#work_dynamic_shapes and https://github.com/NVIDIA/TensorRT/tree/f5e8b8c55060447c4d3a45e7379bc25374edc93f/samples/opensource/trtexec and so on...

from torch2trt.