-
Notifications
You must be signed in to change notification settings - Fork 48
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
File Limit Request: tensorrt - 1000 MB #2609
Comments
#1856 is a past request about this. |
Hey @zeroepoch thanks for the report. Having only one wheel for Linux (Python 3.6+) will reduce your release size quite a lot, have you considered? I suspect that if you continue like you have been doing it, you might reach the project limit in the next release. |
@cmaureir, TensorRT is using pybind11 to build our Python bindings and from what I'm reading in pybind/pybind11#1755 they don't have plans to allow an option or feature to use only the limited API. We do have some plans for next release to reduce the wheel size, maybe using a source package instead that has a downloader inside. Although, for our next release we will not be able to resolve the immediate wheel size issue and will still need to increase the file limit. We understand we need a different solution in the future to avoid the project limit. |
Thanks for your answer @zeroepoch 👍 I've set the upload limit for |
@cmaureir, I forgot that we don't have a TestPyPI project for this module yet. You can ignore that request for now and I'll create a new issue if needed when that time comes. Hopefully it won't. |
Project URL
https://pypi.org/project/tensorrt/
Does this project already exist?
New Limit
1000
Update issue title
Which indexes
PyPI, TestPyPI
About the project
NVIDIA TensorRT has been in development for over 6 years and is now up to version 8.5. We first started with wheels on our own index, but we started pushing wheels to PyPI.org a few releases ago, so users would have an easy time installing TensorRT and other Python projects that depend on TensorRT wouldn't have to go through complications to depend on it.
We don't bundle any files that are outside of our project. We do depend on some other large Python projects which has allowed us to reduce the size of our wheel files some. We have tried to apply more extensive compression, strip more symbols from pre-compiled libraries, but due to new features supported by NVIDIA CUDA 12.0 the size has increased. We are working with other internal library teams to try and reduce our pre-compiled library sizes, but it's not possible without significant development that will take multiple releases.
We do not bundle any sample data in our wheel file and samples are fetched from another location (GitHub).
Reasons for the request
Our last TensorRT release (version 8.5.x) was 550M. Our upcoming release is trending toward 850M, which will exceed the previous limit of 800M that was requested by NVIDIA. We are requesting an increase to the limit so our next release can be pushed successfully. We are aware that 1GB is a hard limit, which is why we are working with internal teams as mentioned above to cut the size down over the next few releases.
Code of Conduct
The text was updated successfully, but these errors were encountered: