Support for using DeepSpeed Layers for inference without going through Hugging face

[bharatv007]

I would like to define my own GPT model architecture and use DeepSpeed layers for inference.

Any tutorials on how to make a GPT model using DeepSpeeds custom layers for inference without goinng through Hugging Face.

Thanks!