Logger sends messages to stdout rather than stderr

[HolyWu]

Description

https://docs.nvidia.com/deeplearning/tensorrt/api/python_api/infer/Core/Logger.html said the log method logs a message to stderr. But the message is sent to stdout instead.

Environment

TensorRT Version: 10.7.0

NVIDIA GPU: GeForce RTX 4060 Ti

NVIDIA Driver Version: 566.36

CUDA Version: 12.6

CUDNN Version:

Operating System: Ubuntu 24.04.1 LTS in WSL2

Python Version (if applicable): 3.12.3

Tensorflow Version (if applicable):

PyTorch Version (if applicable):

Baremetal or Container (if so, version):

Steps To Reproduce

Commands or scripts:

(trt_test) holywu@HOLYWU:~$ pip3 list
Package                  Version
------------------------ -------
nvidia-cuda-runtime-cu12 12.6.77
packaging                24.2
pip                      24.3.1
setuptools               75.6.0
tensorrt                 10.7.0
tensorrt_cu12            10.7.0
tensorrt-cu12-bindings   10.7.0
tensorrt-cu12-libs       10.7.0
wheel                    0.45.1
(trt_test) holywu@HOLYWU:~$ python3 -c "import tensorrt as trt; logger = trt.Logger(); logger.log(trt.Logger.WARNING, 'Hello World')" 2>/dev/null
[12/30/2024-22:40:08] [TRT] [W] Hello World

Have you tried the latest release?: Yes

[pranavm-nvidia]

For the default logger, the documentation is wrong (should say stdout). Custom loggers can obviously log wherever they want to. I will fix the documentation. Thanks for reporting this!

[HolyWu]

Actually I would expect the message was by default sent to stderr instead of stdout. The standard Python libraries warnings and logging also send messages to stderr.

[pranavm-nvidia]

That makes sense, but the default logger has used stdout for years, so probably best not to change it without a strong reason.

It should be quite easy (<10 LOC) to redirect the logs to stderr with a custom logger if you prefer it that way.