Replies: 4 comments
-
Hey, this is the MXNet Label Bot. |
Beta Was this translation helpful? Give feedback.
-
@mxnet-label-bot add [question] |
Beta Was this translation helpful? Give feedback.
-
@frankfliu Thank you for reply! But it seems as failed unit test. have your building systems passed it? Or you don't check tensorrt integration? |
Beta Was this translation helpful? Give feedback.
-
@frankfliu |
Beta Was this translation helpful? Give feedback.
-
Hi, everyone!
I'm got a problem with deficiencies between outputs of model in mxnet and tensorRT. When i start this test https://github.com/KellenSunderland/incubator-mxnet/blob/600bc1f1d7b85ff507ede08715e560722829d837/tests/python/tensorrt/test_resnet18.py, These is an error
AssertionError: Items are not equal: Error 99598.921875 exceeds tolerance rtol=0.000100, atol=0.000100. Location of maximum error:(356,), a=10.381552, b=0.038473 a: array([ 0.36684713, 2.8859494 , -0.544983 , ..., -0.7734007 , 2.931011 , 1.8106198 ], dtype=float32) b: array([0.0387775 , 0.04588229, 0.03329143, ..., 0. , 0. , 0. ], dtype=float32)
How can i synchronize models outputs? At the moment huge drop in quality makes impossible using tensorrt model.
Software installed:
mxnet-cu90 1.4.0.post0
mxnet-tensorrt-cu90 1.3.0
cudnn-7.5
TensorRT-5.1.2.2
Beta Was this translation helpful? Give feedback.
All reactions