This open-source model was created by Nvidia. You can find the release here. The model is available on the huggingface hub: https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF. The 70B model is an RLHF finetuned version of Llama-3.1-70B-Instruct, and supports up to 128K token contexts.