Tensorflow 2.0 & TF-TRT
srkm009 opened this issue · 1 comments
Hi,
Looking forward for the overall support for the TF2.0 workflow since we have finally TF2.0 officially released. Currently, it is failing without any errors for a lot of networks with no uncommon/ custom layers and in successful cases, i dont notice any speedup at all. TF-TRT indeed has a lot of potential for the deployment use-cases. {I have seen the tf-trt working pretty good with TF1.14 etc.}
By workflow, i am referring to :
`from tensorflow.python.compiler.tensorrt import trt_convert as trt
params = trt.DEFAULT_TRT_CONVERSION_PARAMS._replace(
precision_mode='FP16')
converter = trt.TrtGraphConverterV2(input_saved_model_dir=input_saved_model_dir, conversion_params=params)
converter.convert()
converter.save(output_saved_model_dir)`
Related Issue:- #122
TF-TRT is supported in TF2.0.
We don't have many examples yet working in 2.0, but we are working on that and improving the current ones.
I think the image classification example in the experimental branch is in a reasonable shape. Please use it and let us know your opinion.
Any help from the community is appreciated. Thanks.