Why would I spend all that time creating these datasets and training just to brain damage the models with wikitext during quantization? This dataset is primarily multi turn ERP chat. It's formatted to be a drop in replacement for wikitext for quantization methods such as AutoGPTQ or AWQ.
