--- license: mit datasets: - vicgalle/alpaca-gpt4 - BelleGroup/train_1M_CN - stingning/ultrachat - HuggingFaceH4/no_robots - Open-Orca/OpenOrca language: - zh - en pipeline_tag: conversational tags: - Mistral --- # Zephyr-8x7b:Zephyr Models but Mixtral 8x7B We present to you the Zephyr-8x7b, a Mixtral 8x7B MoE model that SFT-only training on a dataset of nearly four million conversation corpora. It has demonstrated strong contextual understanding, reasoning, and human moral alignment without alignment techniques like DPO, and we invite you to participate in our exploration!