[spott]

https://github.com/NVIDIA/Megatron-LM

This is probably a good baseline to start thinking about LLM training at scale.