sparseml icon indicating copy to clipboard operation
sparseml copied to clipboard

onnxruntime-gpu - is there a reson this is not an option?

Open meeki007 opened this issue 2 years ago • 4 comments

Is your feature request related to a problem? Please describe. built the onnxruntime-gpu to fulfill requirement

 Could not find a version that satisfies the requirement onnxruntime>=1.0.0 (from sparseml[torch,torchvision]) (from versions: )
No matching distribution found for onnxruntime>=1.0.0 (from sparseml[torch,torchvision])

Describe the solution you'd like can you make an XOR for onnxruntime >=1.0.0 or onnxruntime-gpu>=1.0.0 in sparseml/setup.py

Additional context should work. git cloning https://github.com/neuralmagic/sparseml.git and going to try and see if it still works after changing line in setup.py to onnxruntime-gpu>=1.0.0 It should as

$ python3 -c 'import onnxruntime as rt; print(rt.__version__)'

gives me

1.12.0

using onnxruntime-gpu I built.

meeki007 avatar Apr 13 '22 14:04 meeki007

Hi @meeki007, thanks for bringing this issue up. We'll dive in and get a solution up soon. In the meantime, let us know if you have any issues with editing the setup.py.

markurtz avatar Apr 13 '22 22:04 markurtz

no issues worked fine using: onnxruntime-gpu>=1.0.0 in setup.py I'm a python nublit but getting better. I lack the skillz to create a PR for this. Give me a few more months in python and I'll be able to help with this simple if than checks stuff.

Oh I did notice that when I installed sparsml:

sudo -H python3 -m pip install -e ./[torch,torchvision]

later when I tried SparseML Ultralytics YOLOv5 Integration I found that it failed checks on the github.com/neuralmagic/yolov5/tree/release/0.12)/requirements.txt

sparseml[torch,torchvision]>=0.12

as it is looking for a package installed via python and not by dev build as I could run sparseml in python and it version checked fine I just removed the requirement along with the requirement for

opencv-python

as I have a full cuda enabled version of openCV built, make, and installed as of right now sparseml is working. going to let it run tonight sparse-ing away

Great work on this. Sorry I'm a JS/BASH/node guy. I look forward to contributing as i can with python as I learn syntax of it.

meeki007 avatar Apr 13 '22 22:04 meeki007

Hey @meeki007. Thank you for your active support! I am working on the issue, the solution is coming soon.

dbogunowicz avatar Apr 22 '22 09:04 dbogunowicz

Hey @meeki007. The new release 0.13 (to be released very soon) should fix the problem. The new version of sparseml shall accept whichever onnxruntime you have previously installed in your environment. Stay tuned!

dbogunowicz avatar May 23 '22 07:05 dbogunowicz

Hi @meeki007 As we're now on release 1.3, the updated versions of sparseml should address your ability to accept previous onnxruntime installations. I will close out this issue but if you find that's not the case, please re-open this issue. Thank you so much! Regards, Jeannie / Neural Magic

jeanniefinks avatar Feb 01 '23 19:02 jeanniefinks