TenyxChat
About
The TenyxChat family is a series of advanced language models developed by Tenyx, designed to function as effective chat assistants. These models are fine-tuned using Tenyx's proprietary Direct Preference Optimization (DPO) framework on open-source AI feedback datasets, such as UltraFeedback. The TenyxChat series includes models like TenyxChat-7B-v1 and Llama3-TenyxChat-70B, which are optimized for multi-turn conversations and exhibit robustness against common fine-tuning challenges like catastrophic forgetting. These models have achieved high rankings on benchmarks such as MT-Bench, often outperforming other models like GPT-3.5 turbo in chat-specific evaluations. The TenyxChat models are available under open-source licenses, making them accessible for further research and commercial use.