This version of TensorRT does not support dynamic ReverseSequence length

spolisetty · October 18, 2021, 4:58pm

Hi,

Regarding primary error you’re facing, currently dynamic dimensions not supported for the ReverseSequence operator.

We tried to build engine using onnx model you’ve shared, but facing following error.
TRT support tensor at most 2G elements.

[10/18/2021-16:55:07] [E] [TRT] ModelImporter.cpp:772: --- End node ---
[10/18/2021-16:55:07] [E] [TRT] ModelImporter.cpp:775: ERROR: ModelImporter.cpp:179 In function parseGraph:
[6] Invalid Node - StatefulPartitionedCall/model/bidirectional/forward_lstm/PartitionedCall/while_loop
[graphShapeAnalyzer.cpp::processCheck::582] Error Code 4: Internal Error ((Unnamed Layer* 60) [LoopOutput]_output: tensor volume exceeds (2^31)-1, dimensions are [2147483647,32,64])
[10/18/2021-16:55:07] [E] Failed to parse onnx file
[10/18/2021-16:55:07] [I] Finish parsing network model
[10/18/2021-16:55:07] [E] Parsing model failed
[10/18/2021-16:55:07] [E] Failed to create engine from model.
[10/18/2021-16:55:07] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v8200] # /usr/src/tensorrt/bin/trtexec --onnx=bi_list_fixed.onnx --verbose --workspace=8000

Following similar issue may help you.

github.com/NVIDIA/TensorRT

How to deal with converting a large onnx model to TensorRT engine?

opened 03:31AM - 07 Jul 21 UTC

closed 07:51AM - 09 Jul 21 UTC

chiehpower

triaged TRT limitation

## Description I have a very large onnx model (almost over 10GB) that I wanna… convert it to TensorRT engine. However, when I converted the onnx model with external data to TensorRT engine, it encountered the error below: ``` &&&& RUNNING TensorRT.sample_onnx # ./sample_onnx DataDirs Area: add the data directory. [07/06/2021-15:44:36] [I] Building and running a GPU inference engine Model path: /workspace/model_external_10GB.trt [07/06/2021-15:44:36] [I] [07/06/2021-15:44:36] [I] >> Start to generate the model. ---------------------------------------------------------------- Input filename: /workspace/model/model.onnx ONNX IR version: 0.0.6 Opset version: 11 Producer name: pytorch Producer version: 1.7 Domain: Model version: 0 Doc string: ---------------------------------------------------------------- [07/06/2021-15:45:09] [W] [TRT] [TRT]/home/jenkins/workspace/OSS/L0_MergeRequest/oss/parsers/onnx/onnx2trt_utils.cpp:220: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32. [07/06/2021-15:45:10] [E] [TRT] Tensor: (Unnamed Layer* 125) [Constant]_output at max batch size of 1 exceeds the maximum element count of 2147483647 [07/06/2021-15:45:10] [E] [TRT] Network validation failed. &&&& FAILED TensorRT.sample_onnx # ./sample_onnx ``` This one was written by C++ TensorRT API to generate the TensorRT model from Onnx. In addition, I can use this code to generate the TensorRT engine with a smaller onnx model with external data (1.1GB), so I think that the code was working well. ## Environment **TensorRT Version**: 7.1 **NVIDIA GPU**: Titan V **NVIDIA Driver Version**: 450 **CUDA Version**: 10.2 **Operating System**: ubuntu 1804 **Python Version (if applicable)**: 3.6 **PyTorch Version (if applicable)**: 1.7 ## Relevant Files Sorry, the onnx model is over 10GB that it is hard to share. I wonder is there anyone who successfully converted a very large model to TensorRT engine before? Is there any relevant document? Thank you so much!!! Best regards, Chieh

Thank you.