Builder.build_cuda_engine(network) silently returns None

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

A clear and concise description of the bug or issue.

Environment

TensorRT Version: 7.2.2.3 (installed in virtualenv with nvidia-tensorrt)
GPU Type: 2070
Nvidia Driver Version: 460
CUDA Version: 11.0
CUDNN Version: 8.0.5.39-1
Operating System + Version: Ubuntu 18.04
**Python Version **: 3.8.8
**PyTorch Version **: 1.7.1

Relevant Files

I am attempting to export and run as a TRT model efficientnet_b2 from this repo:

I export like:
EXPORT = True
output = “my_tf_3.onnx”

if EXPORT:
    import geffnet
    import onnx
    model = geffnet.create_model(
                    'efficientnet_b2',
                    num_classes=3,
                    in_chans=3,
                    pretrained=False,
                    checkpoint_path="/home/luke/projects/traffic_light_classifier/tf_3.pt",
                    exportable=True)
    model.eval()
    DEVICE = torch.device("cuda")
    model.to(DEVICE)
    example_input = torch.randn(1, 3, 256, 256).to(DEVICE)
    model(example_input)
    
    input_names = ["input0"]
    output_names = ["output0"]
    torch.onnx.export(model, example_input, output, verbose=False,
                      input_names=input_names, output_names=output_names,
                      export_params=True)

    print("==> Loading and checking exported model from '{}'".format(output))
    onnx_model = onnx.load(output)
    onnx.checker.check_model(onnx_model)  # assuming throw on error
    print("==> Passed")

A netron diagram of the exported ONNX:

my_tf_3.onnx
my_tf_3.onnx283×24000 1.21 MB

the exported ONNX:
my_tf_3.onnx (29.3 MB)

I try to run as a TRT engine:

import pycuda.driver as cuda
import pycuda.autoinit
import numpy as np
import tensorrt as trt

TRT_LOGGER = trt.Logger()

def build_engine(onnx_file_path):
    builder = trt.Builder(TRT_LOGGER)
    network = builder.create_network()
    parser = trt.OnnxParser(network, TRT_LOGGER)

    builder.max_workspace_size = 1 << 30
    builder.max_batch_size = 1

    # parse ONNX
    with open(onnx_file_path, 'rb') as model:
        print('Beginning ONNX file parsing')
        parser.parse(model.read())
    print('Completed parsing of ONNX file')

    # generate TensorRT engine optimized for the target platform
    print('Building an engine...')
    engine = builder.build_cuda_engine(network)
    context = engine.create_execution_context()
    print("Completed creating Engine")

    return engine, context

engine, context = build_engine("/home/luke/projects/traffic_light_classifier/my_tf_3.onnx")

And I get the error:

 30     print('Building an engine...')
 31     engine = builder.build_cuda_engine(network)
---> 32     context = engine.create_execution_context()
 33     print("Completed creating Engine")
 34 

AttributeError: 'NoneType' object has no attribute 'create_execution_context'

I’m not even really sure how to begin debugging this, any ideas?

1 Like
3

Hi @lukee2ni6,

Looks like engine building is failing. Could you please share complete logs for better debugging.

You can also try generating serialized engine quickly using trtexec command
https://github.com/NVIDIA/TensorRT/tree/master/samples/opensource/trtexec

Thank you.

4

Same problem for me, it works with CLI tetexec but not in python… I got a silent None from builder.build_serialized_network(network, config)

1 Like
5

@trkodpd Hey, I have a similar issue. I have also created this post in case you want to give a follow up.

Would you mind telling me what procedure you followed to run the trtexec script from Nvidia?

1 Like
6

I was able to convert an onnx model to tensorrt engine by using the trtexec command as:

alias trtexec=/usr/src/tensorrt/bin/trtexec
trtexec --onnx=r_pretrained.onnx --saveEngine=r_pretrained.engine --fp16