Sam O

I'm having a similar issue. The inability to specify watchNamespace for multiple namespaces and the inability to create multiple deployments means that it is impossible to deploy load balancers for...

Hey there! I have an approved MR [here](https://github.com/NVIDIA/NeMo/pull/13367) that tackles a similar problem but does it slightly differently. We are using the {% generation %} tags from the template and...

@jenchen13 rebase completed, functionality is merged. Only changed the default value in GPTSFTChatDataset and not in ChatDatamodule so that the behavior from the outer interface didn't change.

For reference the ChatDataModule that we use in Customizer is [here](https://gitlab-master.nvidia.com/aire/microservices/nmp/-/blob/main/services/customizer/src/training/nemo/datamodules/chat_enabled_datamodule.py?ref_type=heads), we base off of FinetuningDatamodule and make it compatible with sequence packing.

+1 Some mechanism for indicating when layers are evicted would be greatly appreciated.