They do mention the 8b+ "GPT-2" model trained by nVidia. Which is their reference to megatron.