[request] onnxruntime/1.13.1 #16699
Comments
|
duplicate of #4806 |
I have managed to reproduce results from @CAMOBAP on Mac / Linux / Windows (with some bug fixes) But the main problem with this package is that they using subprojects to random commits (looks like). And that is why I think it will be many problems to build such new version, but, you can close it if needed, I will up #4806 |
|
@elejke I have a partial but working recipe of onnxruntime 1.14.1 here: I wrote this before I noticed #4086 and it's not nearly complete as that effort (very few options, just tried to get CPU engine working). However, I mostly got through the major cmake changes in recent onnxruntime releases. They started using FetchPackage in cmake 3.24 so the original recipe is out of date (onnxruntime_PREFER_SYSTEM_LIB is gone). If you're willing to take this over, that'd be great. This has reached the limit of my knowledge of cmake, and I'm not available at the moment. It's using the new cmake tool. The dependencies are mostly solved except:
The patches are quite extensive. Not sure how to handle that. Maybe some of it can get merged upstream in onnxruntime, like the package name calls to FetchPackage. |
|
for the start: #16701 🙂 great work! I have been disappointed when discovered recent changes in ort building workflow and decided that reproducing the 1.7.1 results are almost nothing to make it work with >=1.13.1, but your recipe looks promising!! did you check it on all desktop os? @CAMOBAP's recipe works for cross-building from linux to all android abi's after small fixes don't you want to create a PR with it and let others to help to push it to cci finally? |
|
Okay, I've put up #16842 but it could use a lot of help. I've tested macOS, Windows, and Linux, but not iOS or Android, and of course all the options for execution providers that should go in there. |
…LLY_DISCONNECTED=ON` (#15323) ### Description Rework some external targets to ease building with `-DFETCHCONTENT_FULLY_DISCONNECTED=ON` This will allow package managers to more easily provide an onnxruntime package by reducing the amount of patching needed downstream at each version. ### Motivation and Context Availability of onnxruntime in some C++ package managers #7150 conan-io/conan-center-index#16699 microsoft/vcpkg#20548 My initial intent is to get this in conan but the PR would most likely be useful (though not tested) to vcpkg as well (and maybe others). I tried to get only a first batch of not too specific patches (i.e. not specific to conan). The first commit reworks `flatbuffers` and just extends what @snnn did in #13991 The second commit reworks `pytorch_cpuinfo` The third commit reworks `google_nsync`
…LLY_DISCONNECTED=ON` (microsoft#15323) ### Description Rework some external targets to ease building with `-DFETCHCONTENT_FULLY_DISCONNECTED=ON` This will allow package managers to more easily provide an onnxruntime package by reducing the amount of patching needed downstream at each version. ### Motivation and Context Availability of onnxruntime in some C++ package managers microsoft#7150 conan-io/conan-center-index#16699 microsoft/vcpkg#20548 My initial intent is to get this in conan but the PR would most likely be useful (though not tested) to vcpkg as well (and maybe others). I tried to get only a first batch of not too specific patches (i.e. not specific to conan). The first commit reworks `flatbuffers` and just extends what @snnn did in microsoft#13991 The second commit reworks `pytorch_cpuinfo` The third commit reworks `google_nsync`
Package Name/Version
onnxruntime/1.13.1
Webpage
https://www.onnxruntime.ai/
Source code
https://github.com/microsoft/onnxruntime
Description of the library/tool
ONNX Runtime is a cross-platform inference and training machine-learning accelerator compatible with deep learning frameworks, PyTorch and TensorFlow/Keras, as well as classical machine learning libraries such as scikit-learn, and more.
There was one good try to create a recipe from @CAMOBAP, even I managed to build it now for Mac OS for version 1.7.1 and test_package passes correctly.
On Windows it have some errors, but it looks like there was a solution for them before from @fdgStilla (windows wil library should be added as build_requirement)
As for now 1.7.1 is too old and it will be perfect to update the recipe and finally get it to the conan center.
This is the most popular library for neural network inference on all platforms!
The text was updated successfully, but these errors were encountered: