sali1997s

Multiple sharding policy in plugin.jax.data_iterator

3

### Describe the question. Is there a way to set jax sharding for each output separately plugin.jax.data_iterator? For example, I have pipeline, that has 2 outputs. I want first output...

- Docker: nvcr.io/nvidia/jax:23.10-t5x-py3 - GPU/TPU model and memory: 8 A100 - 40Gb I think, there is a trouble with grouped convolution when i'm using distributed training. I have tried to...