TensorRT icon indicating copy to clipboard operation
TensorRT copied to clipboard

Published 20 hours ago • verified publisher icon NVIDIA

Missing required positional arguments calling _output_padding in _ConvTransposeNd from torch.

Open geoffrey-g-delhomme opened this issue 1 year ago • 5 comments

Fix #2964

geoffrey-g-delhomme avatar May 12 '23 20:05 geoffrey-g-delhomme

/blossom-ci

samurdhikaru avatar Jun 13 '23 20:06 samurdhikaru

/blossom-ci

samurdhikaru avatar Jun 13 '23 20:06 samurdhikaru

/blossom-ci

samurdhikaru avatar Jun 13 '23 20:06 samurdhikaru

thanks for the contributing, the latest code is in branch release/9.1, and for compatibility, we inspect the signature first to see if those spatial dims and dilations are needed. see:

https://github.com/NVIDIA/TensorRT/blob/release/9.1/tools/pytorch-quantization/pytorch_quantization/nn/modules/quant_conv.py#L278

ttyio avatar Nov 29 '23 17:11 ttyio

When will the pytorch quantification package be updated synchronously for this issue? It's too troublesome to modify the source code every time.

yjszyd0813 avatar Feb 19 '24 07:02 yjszyd0813