preparing sync between branches to avoid divergence #330
Conversation
regular sync
regular sync
dev <- Mlperf inference
Main <- Dev
* added more standard Nvidia Docker configuration for PyTorch
* renamed dummy script to test-dummy
…le already exists
Update customize.py | Fix download-file on windows when downloaded fi…
various updates for CM tools for Windows and better debugging
turn on tests on Windows
* added windows test for image classification
* removed windows test for MLPerf (requires interaction)
Fixes in download-file script for windows
various updates for CM tools for Windows and better debugging
fixing a few outdated URLs for Windows
fixing a few outdated URLs
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
| ## CM concepts | ||
|
|
||
| * https://doi.org/10.5281/zenodo.8105339 | ||
| * https://arxiv.org/abs/2406.16791 |
There was a problem hiding this comment.
We should remove this arxiv paper - it has many factually wrong things. You had earlier agreed to fix them.
There was a problem hiding this comment.
?
It has a few minor typos that I still need to clean up but it was not urgent. But if you feel that something is wrong, I will remove it for now to discuss it later. There are also many other things I would like to clarify with the very recent CM/CM4MLOps developments while I was on a sick leave before updating the paper.
There was a problem hiding this comment.
Here is the PR: #336
There was a problem hiding this comment.
Thank you Grigori. Yes, not just the typos - "Arjun Suresh - a senior software engineer from cKnowledge.org" - I was at OctoML and only later at cKnowledge and even there the title was not the same. And I think in the last section "Future Plans" you made a typo to use "We" as everywhere else in the PDF it is "I".
| @@ -1,5 +1,5 @@ | |||
| Copyright (c) 2021-2024 MLCommons | |||
|
|
|||
| The cTuning foundation and OctoML donated this project to MLCommons to benefit everyone. | |||
| Grigori Fursin, the cTuning foundation and OctoML donated this project to MLCommons to benefit everyone. | |||
There was a problem hiding this comment.
We should agree on the history and keep it stable - it doesn't look good if it keeps on changing. Technically cm4mlops and all the code inside it started within MLCommons.
There was a problem hiding this comment.
Parts of the code comes from older versions of CK and CM and from the https://github.com/mlcommons/ck where it was originally resided. We have to reflect that to avoid legal problems in the future.
There was a problem hiding this comment.
I originally started developing CM4MLOps as an integral part of CM while reusing some parts of CK and CK4MLOps that I developed at cTuning so it has to have a common history and license.
There was a problem hiding this comment.
https://octo.ai/blog/octoml-joins-the-community-effort-to-democratize-mlperf-inference-benchmarking - CM and CM4MLOps was a part of CK project at that time.
There was a problem hiding this comment.
The OctoAI blog has nothing to do with CM as it is talking about results taken using CK. Anyway I do not want to argue over these things. Since none of the projects are new, I request you to finalize the historical attributions and contributions and and keep them constant. This can help others decide in using the project. If anyone has a concern that the history of a project can change in future no one will be happy to contribute to it. Or else, let git history take care of them automatically as done in most other MLCommons repositories.
There was a problem hiding this comment.
Sure! Agree! Thank you!
But we still have a problem with the conflict in .github/workflows/test-mlperf-inference-resnet50.yml .
May I ask you to help resolve it and merge the PR, please?
Thanks a lot for your help - very appreciated!
|
@anandhu-eng @arjunsuresh - I don't understand why but the button to "Resolve conflicts" is gray and I have a problem resolving a small conflict in 1 file before merging this PR... Do you think you can help me resolve the conflict in .github/workflows/test-mlperf-inference-resnet50.yml, please? Thank you very much in advance! |
As discussed (#327 , #257), let's start syncing branches to avoid further divergence and have time to test current workflows from main & mlperf-inference before changing json to yaml ...
@anandhu-eng and @arjunsuresh - can you please check how it influences the mlperf-inference branch, please?
BTW, I somehow thought that we synced all branches about a month ago so I am not sure why there is such a difference ...
The code should be mostly stable in all branches so it should be possible to check and test changes soon.