Yi-Lightning Technical Report

TL;DR

Yi-Lightning is a new large language model that achieves top-tier performance across various tasks using an advanced MoE architecture, multi-stage training, safety measures, and scalable infrastructure, while highlighting limitations of traditional benchmarks.

Contribution

The paper introduces Yi-Lightning, a large language model with innovative architecture, training strategies, safety framework, and insights into benchmark limitations, advancing practical AI development.

Findings

Achieves 6th place on Chatbot Arena, with top results in Chinese, Math, Coding, and Hard Prompts.

Utilizes an enhanced MoE architecture with optimized routing and KV-caching.

Demonstrates competitive performance on public benchmarks and reveals disparities between static benchmarks and human preferences.

Abstract

This technical report presents Yi-Lightning, our latest flagship large language model (LLM). It achieves exceptional performance, ranking 6th overall on Chatbot Arena, with particularly strong results (2nd to 4th place) in specialized categories including Chinese, Math, Coding, and Hard Prompts. Yi-Lightning leverages an enhanced Mixture-of-Experts (MoE) architecture, featuring advanced expert segmentation and routing mechanisms coupled with optimized KV-caching techniques. Our development process encompasses comprehensive pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF), where we devise deliberate strategies for multi-stage training, synthetic data construction, and reward modeling. Furthermore, we implement RAISE (Responsible AI Safety Engine), a four-component framework to address safety issues across pre-training, post-training, and serving phases. Empowered by our scalable super-computing infrastructure, all these innovations substantially reduce training, deployment and inference costs while maintaining high-performance standards. With further evaluations on public academic benchmarks, Yi-Lightning demonstrates competitive performance against top-tier LLMs, while we observe a notable disparity between traditional, static benchmark results and real-world, dynamic human preferences. This observation prompts a critical reassessment of conventional benchmarks' utility in guiding the development of more intelligent and powerful AI systems for practical applications. Yi-Lightning is now available through our developer platform at https://platform.lingyiwanwu.com.