Shenzhen Yuanxiang Information Technology Co., Ltd. recently announced the successful release of China’s largest Mixture of Experts (MoE) open source large model — XVERSE-MoE-A36B. The launch of this model marks a significant advancement in China’s AI field, elevating domestic open source technology to an internationally leading level.

The Yuanxiang Open Source High-Performance Large Model XVERSE-65B-2 Base Version focuses on enhancing code and math capabilities. XVERSE-65B-2 has been optimized in Continual Pre-Training, with a total training token volume of 3.2 trillion. Yuanxiang's large model surpasses GPT-3.5 and Llama2-70B in comprehensive evaluations, and it significantly exceeds open-source benchmarks. XVERSE-65B performs exceptionally well in SuperCLUE Chinese general large model tasks.

YuanXiang announced the open-source high-performance large model XVERSE-65B-Chat edition, which is unconditionally free for commercial use. XVERSE-65B ranks first in the national open-source total score in the SuperCLUE comprehensive benchmark assessment for Chinese general large models. XVERSE-65B has stronger understanding, generation, logic, and memory capabilities, capable of handling more diverse and challenging tasks. YuanXiang has open-sourced three large models with 7 billion, 13 billion, and 65 billion parameters, among which XVERSE-65B is one.

Yuanxiang Information Technology announces the launch of multiple high-performance models of the Yuanxiang large model, all open-sourced and free for commercial use. This fills the gap in the Chinese large model open-source ecosystem and provides technical services to help achieve industry-leading capabilities in excellent Chinese creation, logical reasoning, and task execution.

Yuanxiang XVERSE has partnered with Tencent Music to launch the lyraXVERSE accelerated large model, which is applied in the music assistant 'AI Xiaoqin', providing a personalized music interaction experience. Users can interact with AI Xiaoqin through QQ Music, including casual chatting, music recommendations, music interpretation, and songwriting. The lyraXVERSE large model enhances the emotional and intellectual capacity of the AI assistant, offering deeper music interactions. The large model application improves operational efficiency and user experience, increasing response accuracy and the depth of user conversations.

Yuanxiang open sources the 70 billion parameter general model XVERSE-7B, fully open source and free for commercial use, supporting over 40 languages. XVERSE-7B can run inference on a single consumer-grade GPU with a minimum of only 6GB of VRAM after quantization.