TensorRT

TensorRT copied to clipboard

## Bug Description Got nan result if batchnorm constant fold is enabled in RTX if disable the batchnorm constant fold in RTX, it is working as expected. ## To Reproduce...

lanluo-nvidia

Logs: ``` File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 524, in __call__ return wrapper() ^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/_ops.py", line 515, in wrapper return torch.overrides.handle_torch_function( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/dist-packages/torch/overrides.py", line 1725, in handle_torch_function result = mode.__torch_function__(public_api, types,...

cehongwang

# Description Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change....

cehongwang

## TL;DR Based on the PR #3732, converters need to be optimized to meet the requirement of converter unit tests that Torch-TRT is expected to have the same num of...

zewenli98

## Bug Description Torch-TRT has more Reformatting than ONNX-TRT, which affects perf. For example: ``` , { "name" : "Reformatting CopyNode for Input Tensor 1 to [CONVOLUTION]-[aten_ops.convolution.default]-[model.0.residual/convolution] + [ELEMENTWISE]-[aten_ops.add.Tensor]-[model.0/add]", "timeMs"...

zewenli98