Adam Pocock

You're writing the data in channels last format `[width,height,channels]` but YOLO5 wants channels first `[channels,width,height]`. So your image is corrupted when YOLO5 sees it. You need to change how you...

You're writing out the elements with the channels in the last dimension, but you then construct the tensor telling it the channels are the first dimension. It can't do the...

@karllessard should we move to Ubuntu 22.04? 24.04 is going to break a lot of people as it will have a much newer glibc.

The TF-Java API docs are hosted here - https://www.tensorflow.org/jvm, though they are currently quite old and we are trying to update them.

The reference op has it returning the same value each time if the seed is set, and if the seed is not set then it seeds itself from the clock...

The ONNX Runtime behaviour with a stateful generator inside each instance of the op is fine, though it's a bit of a pain to track the RNG state if you...

My issue with the seeded vs unseeded behaviour in the reference is not determinism vs non-determinism, it's that the seeded one always produces an identical output each time the op...

You can't use kotlinx.serialization directly, we don't annotate our classes with its serializable interface. However you can serialize Tribuo types to protobufs, then pull out the byte array from the...

TensorFlow will optimize things based on the available CPU instructions, so if you have Intel Xeons with AVX-512 and older AMD Epycs without AVX-512 then you'll get a lot faster...

After downloading the tar file this was sufficient for me to export it: ```python import keras_hub gemma_lm = keras_hub.models.GemmaCausalLM.from_preset("gemma2-keras-gemma2_instruct_2b_en-v1") gemma_lm.export("gemma2-2b-it-saved-model") ``` The saved model is massive (~19GB) for some reason...