Dolphin Mixtral 8x7B

Dolphin Mixtral 8x7B is an uncensored, instruction-tuned mixture-of-experts model built on Mistral's Mixtral architecture. With 47 billion total parameters and approximately 12 billion active per inference, it delivers strong performance across text generation, coding, and multilingual tasks without alignment restrictions. This model combines the powerful MoE architecture of Mixtral with Dolphin's uncensored training approach, making it a popular choice for researchers and developers who need high-quality outputs without content filtering. It requires significant VRAM but rewards users with excellent reasoning and creative writing capabilities.