TensorRT icon indicating copy to clipboard operation
TensorRT copied to clipboard

XXX failure of TensorRT X.Y when running XXX on GPU XXX

Open ndeep27 opened this issue 1 year ago • 1 comments

Description

Getting below error when running a TensorRT converted model. 2024-07-08 17:23:06.193422: E external/org_tensorflow/tensorflow/compiler/tf2tensorrt/utils/trt_logger.cc:87] DefaultLogger 7: [shapeMachine.cpp::executeContinuation::864] Error Code 7: Internal Error (IShuffleLayer (Unnamed Layer* 190) [Shuffle]: reshaping failed for tensor: TensorRTInputPH_38 reshape would change volume 97200 to 98100 Instruction: RESHAPE_ZERO_IS_PLACEHOLDER{3240 1 30} {545 180}.)

My original model is in TF and I used the TFTConverter to convert the model

converter = trt.TrtGraphConverterV2( input_saved_model_dir=SAVED_MODEL_PATH, use_dynamic_shape=True, dynamic_shape_profile_strategy="Range", precision_mode=trt.TrtPrecisionMode.FP32 )

trt_func = converter.convert()

Any pointers on the above error? When running the TF model on its own, i don't see any issues in inference.

ndeep27 avatar Jul 08 '24 22:07 ndeep27

similar case https://forums.developer.nvidia.com/t/ishufflelayer-reshape-8-reshaping-failed-for-tensor/277844

lix19937 avatar Jul 09 '24 05:07 lix19937

Seems like your converted ONNX graph has some issue in a reshape node. Could you attach the ONNX model file?

LeoZDong avatar Feb 12 '25 00:02 LeoZDong

Closing due to inactive. Please feel free to reopen!

poweiw avatar May 29 '25 21:05 poweiw