Closed thewh1teagle closed 1 month ago
You'll need to compile ONNX Runtime from source with --use_extensions
.
You'll need to compile ONNX Runtime from source with
--use_extensions
.
I'll try to compile it from source and enable it. Why do you prefer not to enable it? I see that also onnxruntime official builds doesn't include it and I'm wondering why. It has great potential in simplify models usage. for example with whisper I don't need encoder / decoder. single model does it all.
Why do you prefer not to enable it?
Partially because the official builds don't. I don't want to ship builds of ONNX Runtime with any more features than what most developers will use; that would just waste downstream users' disk space. Bandwidth is also a concern (currently I'm fortunate enough to not be billed for egress bandwidth, but as ort
grows this will likely change).
This is the 2nd issue in the history of this project mentioning onnxruntime-extensions
(the first being the PR adding with_extensions
), so I do not currently think the need for it is enough to warrant the cost on my end.
Maybe you could host the binaries in Github releases for free without limits. Thanks for explain.
Hey!
I'm trying to use model which requires onnx runtime extensions with ort. However when using
with_extensions()
it throw error.I'm trying to replicate the following: main.py
Error
Code
OS: macOS m1 ort version:
2.0.0-rc.4