Adam Pocock

I loaded it in in jshell, but my laptop has 48GB of RAM. The model should take about 8GB for the parameters in fp32, I'm not sure why it's 19GB....

We did have an effort a few years ago to support Keras's model formats, but we don't have enough contributors currently to complete that and Keras changed its file format...

The Python snippet isn't applied iteratively, `data` is not fed back in to `data_tensor`. Is this a typo? The `SavedModelBundle` has it's own session & graph so you shouldn't need...

> > Other than the above points, there's also that Java won't warm up in 10 iterations, it'll be more like a few hundred, but you're likely measuring something problematic...

Ok. Did you check the saved model speed in Python, if that's slower than the tf.function and similar to the TF-Java speed then you'd have to look into turning on...

We modified where it's looking for the headers just before the rc1 release to fix this kind of issue. I tested it on macOS, and I thought I had tested...

So it looks like the problem is that we used to get the absl headers from Bazel, but something has changed in the TF build process so it's not putting...

I can replicate this, but we couldn't replicate it on Karl's machine, even after a clean of bazel. Both machines are running macOS 14.5 with the latest XCode, and the...

No, in some cases the external folder in `bazel-tensorflow-core-native` has extra folders in it linking to the dependencies we need the headers for, which we add to the include path...

TF-Java doesn't support h5 files. If you load the model in in TF Python then save it as a TF SavedModel then it should work.