onnxruntime v1.19.2

[regro-cf-autotick-bot]

Closes #131
Closes #133

It is very likely that the current package version for this feedstock is out of date.

Checklist before merging this PR:

• Dependencies have been updated if changed: see upstream
• Tests have passed
• Updated license if changed and license_file is packaged

Information about this PR:

1. Feel free to push to the bot's branch to update this PR if needed.
2. The bot will almost always only open one PR per version.
3. The bot will stop issuing PRs if more than 3 version bump PRs generated by the bot are open. If you don't want to package a particular version please close the PR.
4. If you want these PRs to be merged automatically, make an issue with @conda-forge-admin,please add bot automerge in the title and merge the resulting PR. This command will add our bot automerge feature to your feedstock.
5. If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase @conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

Pending Dependency Version Updates

Here is a list of all the pending dependency version updates for this repo. Please double check all dependencies before merging.

Name	Upstream Version	Current Version
cudnn	9.3.0.75
protobuf	28.0
setuptools	74.1.1

_{This PR was created by the regro-cf-autotick-bot. The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. Feel free to drop us a line if there are any issues! This PR was generated by - please use this URL for debugging.}

[conda-forge-webservices]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

[hmaarrfk]

The file seems there, but maybe just not found:

$ find -name host_config.h
./_h_env/targets/x86_64-linux/include/host_config.h
./_build_env/targets/x86_64-linux/include/host_config.h
./_build_env/targets/x86_64-linux/include/crt/host_config.h

[cbourjau]

Only the Cuda builds are failing now, and I'm a little puzzled by this. The Cuda 11 builds fail with a very similar but not identical error.

[hmaarrfk]

cc: @jakirkham any ideas?

[cbourjau]

Just bumping this for visibility. I could imagine that it is a pretty straightforward fix, but I don't have the hardware to debug this easily. Any help is much appreciated!

[hmaarrfk]

unfortunately, even looking at the cuda 12 i can't figure it out on linux.

[hmaarrfk]

I think jakirkham was AFK last week, so maybe we can ping him again next?

[hmaarrfk]

Just going through some findings;

Th ediff between v1.19.2 shows that we still might need to disable installing requirements on windows.

         if args.enable_pybind and is_windows():
-            install_python_deps(args.numpy_version)
+            run_subprocess(
+                [sys.executable, "-m", "pip", "install", "-r", "requirements/pybind/requirements.txt"],
+                cwd=SCRIPT_DIR,
+            )

[hmaarrfk]

oh seems like you fixed pip for windows great!

[hmaarrfk]

@cbourjau just to confirm, you don't need a CUDA enabled machine on linux, just linux+docker.

I give up for today, it just doesn't seem like their build scripts changed all that much, and the previous build is found fine....

[cbourjau]

I'm afraid I won't have much time to spend on this in the next couple of weeks. Do you think it may be reasonable to temporarily disable the Cuda builds and to release the CPU-only 1.19.2 packages, @hmaarrfk ?

[hmaarrfk]

Do you think it may be reasonable to temporarily disable the Cuda builds

Is this really what you want? Generally speaking CUDA is a great enabling technology for ML.

Do you want to field the slew of questions that will come from CUDA users updating to 19.2 with CPU only support?

I would love it if instead you:

• Enabled upload of artifacts store_build_artifacts: true https://conda-forge.org/docs/maintainer/conda_forge_yml/#azure
• rerendered
• Downloaded them
• Uploaded them to your own channel

Generally this is what I would suggest for others that have "incomplete" packages.

However, I do understand that CUDA is alot of work, but the performance loss so great that it is almost worst to have a onnx package without cuda support......

I am however unable to help for the last few weeks so I can just as easily limit our version of onnx to 18..... (on my own private channel)

[jakirkham]

I think jakirkham was AFK last week, so maybe we can ping him again next?

Yeah was on PTO for a bit

Noticed that upstream made both a 1.18.2 and a 1.19.2 release around the same time

Given this is on 1.18.1, would it be worth trying 1.18.2? This might be a smaller step with fewer changes. Also it might give us the opportunity to fix a few issues before jumping to the 1.19.x series

[hmaarrfk]

Part of the sad sad thing that made me sad is that even rerendering failed on windows:
#131

[jakirkham]

It's looking for nvcc in the host environment instead of the build environment

Please see this CI job with snippet below:

CMake Error at CMakeLists.txt:735 (enable_language):
  The CMAKE_CUDA_COMPILER:

    C:/bld/onnxruntime_1725973208545/_h_env/Library/bin/nvcc

  is not a full path to an existing compiler tool.

Compare this to where it finds cmake

--   CMake command                     : C:/bld/onnxruntime_1725973208545/_build_env/Library/bin/cmake.exe

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found some lint.

Here's what I've got...

For recipe/meta.yaml:

• Recipe maintainer "jtilly" does not exist

[hmaarrfk]

@traversaro you'll be added on the next rerender. I'm trying to cut the cuda architectures, to get it to work on the CIs..... but..... we are now at 80,86,90 which is likely too few even for my taste.

Are you able to build out the windows matrix?

[conda-forge-admin]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe/meta.yaml) and found it was in an excellent condition.

[hmaarrfk]

@conda-forge-admin please rerender

[traversaro]

Are you able to build out the windows matrix?

I can look into that, but it will probably take me a few days. If this is blocking, probably we can skip Windows for the time being? By skipping the whole Windows build (instead of just CUDA) we avoid the problem of people updating and ending up with CPU-only onnxruntime.

[hmaarrfk]

I wonder if we can agree on a smaller build matrix for CFEP03. Historically that hasn’t been favored due to confusion.

I for example only use python 3.10. Looking into the “even versions” of python due to the more frequent releases. I would be fine with python 3.10 and 3.12 only.

[hmaarrfk]

I would to ensure that CFEP03 is like:

Just "run a single command" come back "in 6, 12, 24, 48, 72hrs, doesn't really matter" and upload the jobs.

I'm trying to see if we can implement a "megabuild strategy"

[hmaarrfk]

@conda-forge-admin please rerender

[hmaarrfk]

Well. Very nice. Everything seems borkes

[hmaarrfk]

seems like the bot has trouble rerendering.

[hmaarrfk]

Windows logs for cuda 12.0
win_cuda12_python3XX.zip

packages

• https://anaconda.org/mark.harfouche/onnxruntime-cpp
• https://anaconda.org/mark.harfouche/onnxruntime

Let me know if you are all OK with this.

[hmaarrfk]

sorry about that, the thumbs up didn't trigger a notification. will upload.

[traversaro]

sorry about that, the thumbs up didn't trigger a notification. will upload.

My bad, I am well aware of this, and I did not provide a comment. Thanks a lot for handling the Windows builds here.

[hmaarrfk]

its ok, just explaining the delay.