Our first commercial plugin using ONNX

[mikesolomon]

+1, the subscription idea was based on the fact that we’re always either training the model more (servers in the cloud) or working with new singers to get new data, which costs studio & singer time. The fact that something is on a server definitely winds up being a huge cost for companies like OpenAI, but at our scale, it’s the continuous retraining and data sourcing that represents the largest recurring cost.

We’ve seen that there’s no demand for a subscription, though, and we’ve cut down our retraining budget & our recording budget as a result. Folks seem happy enough with the current plugin, and I think the marginal differences we’re seeing as we tweak it don’t really justify a subscription. Which is not to say we won’t have an epiphany at some point that leads to a major upgrade, but if/when this does happen, it’ll be a fresh angle rather than the incremental approach we’ve been using to retrain the plugin.

[7sharp9]

Thanks for providing the info for the C++ part of things, what was the process with cmake did you statically link. I wondered if you could share any gists from the cmake side of things?

[jonata]

Our first commercial plugin using ONNX

To my knowledge, onnxruntime is not really intended to be built as a static library. I remember trying to get that setup working for quite some time but we finally decided to use a dynamically linked build of onnxruntime. This thread has some information on possible ways to deploy dynamic libraries on windows, especially the latest post describes an interesting option that I’d consider if I had to do it again from scratch today. We went the system32 route and put our custom onnxruntime build the…

Hi @PluginPenguin,
I’m researching this after some issue I had with onnxruntime.dll for my windows users. How can you create a custom dll name and make sure the plugin will load that file on windows?
I assume something with the linking process of the library in CMakeLists.txt, but I coudln’t find how.

Will appreciate your tips for that.

[PluginPenguin]

We actually do this with a bunch of python scripts. First of all, we build our customized version of onnxruntime using the ci_build/build.py script from the onnxruntime repository. Then we rename the resulting dlls using this rename_dll.py script.

Then the plugin itself links against the dlls with the new custom names rather than plain onnxruntime etc.

I don’t know if you also build onnxruntime from scratch, if not I would expect that renaming a prebuilt onnxruntime dll would also work well.

[jonata]

@PluginPenguin

I don’t know if you also build onnxruntime from scratch, if not I would expect that renaming a prebuilt onnxruntime dll would also work well.

thanks, will give it a go with the prebuilt onnxruntime.
For the linking process, I’m just using CMake find_package and giving it the path to the lib directory in the onnxruntime release for windows. If the dll is renamed in this directory is this expected to work?

[fbecir]

Hello

You can also link statically ONNXRuntime :
https://github.com/csukuangfj/onnxruntime-libs

Or you can see an example here : https://github.com/IGNF/JUCE_FLAIR

Hope that helps !

[jonata]

Our first commercial plugin using ONNX

We actually do this with a bunch of python scripts. First of all, we build our customized version of onnxruntime using the ci_build/build.py script from the onnxruntime repository. Then we rename the resulting dlls using this rename_dll.py script. Then the plugin itself links against the dlls with the new custom names rather than plain onnxruntime etc. I don’t know if you also build onnxruntime from scratch, if not I would expect that renaming a prebuilt onnxruntime dll would also work well.

rename_dll is a life save. thank you!