Adam Pocock

Depending on how you've loaded your model in and how it's specified, then you might be able to feed in a boolean (scalar) tensor that specifies if it's training or...

You should use `TBool.class` rather than `Boolean.class`. However I'm not sure that that will achieve what you want. The effect of the `training=true` flag in Keras depends on the layers...

The feeding logic is pretty much exactly how it worked in tf 1. We're working on function support which will line up with how tf.function works in tf 2, this...

We're working on simplifying the build process so we can pull in the TF binaries from the Python whls rather than building it from source. Once that's done it'll be...

We're discussing how to implement it at the moment, so we don't have code yet. Spent last week checking out the whl folder layouts to make sure they all look...

This is pretty interesting. I've been thinking about a similar thing for Java, as specifying the ops by hand in an enum is tedious (which is what we've been doing...

Our current goals are to write ONNX files from Java, but we're also considering using our ONNX op definitions as an eager mode shim similar to [ONNX Script](https://github.com/microsoft/onnx-script) as some...

Sure, I was talking about one of our use cases which is to patch up deficiencies in the tensor libraries available on the JVM by using ops from ONNX when...

Operations for combining ndarrays are available in TensorFlow-Java, either using the imperative eager mode or the more declarative graph mode.

Did Maven download `tensorflow-core-api-0.3.1-windows-x86_64.jar` to your local Maven cache? That jar should contain `jnitensorflow.dll` and `tensorflow_cc.dll`. It's available here - https://repo1.maven.org/maven2/org/tensorflow/tensorflow-core-api/0.3.1/.