Ye Huang

Will provide a minimal code (in colab) in next week

Same. Could you please provide ablation stuides that only use GPU/CPU on iPhone? Depending on the iPhone runtime, when NPU (ANE) is enabled, some operations are not supported and may...

Agree. We need a flag or a function to enable/disable the stateless/stateful rng_type

> Thanks for reporting this issue. As a short cut for now (to have a determined initializer for your test), you can use `tf.keras.utils.set_random_seed()` to set the seed for python/numpy/tf/keras....

@qlzh727 I think providing a flag for the switch between stateless and stateful is still necessary. I found the new RNG behavior (same seed, same hardware with deterministic enabled) produced...

> Same issue here! Using 4GPU for distributed training on Ubuntu 22.04 with Tensorflow 2.13 hangs at the “compiled cluster using the XLA” line. Issue solved by downgrading to 2.12...

> > Same issue here! Using 4GPU for distributed training on Ubuntu 22.04 with Tensorflow 2.13 hangs at the “compiled cluster using the XLA” line. Issue solved by downgrading to...

Upgrade the NVIDIA driver >= 545 and the issue should be addressed

@bhack Since JAX already has Pallas to write the TPU kernel, is there any plan for TensorFlow for a similar feature?

> @edwardyehuang are you interested in contributing this to KerasCV? If MobileNetV4 releases weights, I will be happy to provide KerasCV implementation. I wrote to them a few months ago,...