Text this: SimpleScale: Simplifying the Training of an LLM Model Using 1024 GPUs