Adam Pocock
Adam Pocock
If you're happy for it not to be themed as well as the TF documentation is, and Google is fine with us running a GitHub Pages instance out of the...
Spring AI wants to load it from the classpath so it doesn't have a location on disk. Is there a spec for the `onnx_data` file format anywhere? I can add...
But what's the format of the bytes? Do they have a shape array first then bytes or just bytes? Is there a marker between the tensors?
Ok, thanks. Are the floats stored little endian like protobuf, or is it platform dependent?
That might be a bug in the ORT create buffer logic. Try doing `byteBuffer.order(ByteOrder.LITTLE_ENDIAN).asFloatBuffer()` and then passing that to the `OnnxTensor.createTensor(OrtEnvironment, FloatBuffer, long[])` method. I'll try to run down the...
There is a bug in the create buffer logic, I'll get it fixed next week.
Fix is in this PR - https://github.com/microsoft/onnxruntime/pull/21774.
We should fix that, however it's not clear that upstream TensorFlow is itself covered by the API stability guarantees anymore given all the changes that have happened over there.
Are you mixing TF-Java versions? 1.0.0 doesn't run on Java 8, and you really shouldn't mix tensorflow-core-api 0.4.2 with tensorflow-core-native 1.0.0-rc.2.
We dropped support for Java 8 with TF-Java 0.5.0. That version and 1.0.0-rc.2 require Java 11. You can probably compile the v1.0.0 tag with Java 8 by changing the pom...