Yi-Lightning Technical Report

2024-12-02

arXiv.org

AI Summary

To address the performance–cost imbalance of large language models in Chinese understanding, mathematical reasoning, code generation, and complex prompt handling, this work introduces Yi-Lightning—a high-performance open-source model. Methodologically, it proposes a novel enhanced Mixture-of-Experts (MoE) architecture with dynamic expert routing; designs the RAISE safety framework—comprising Rule-based filtering, Alignment fine-tuning, Iterative refinement, and Safety Evaluation; and integrates synthetic data construction, multi-stage training (pretraining, supervised fine-tuning, RLHF), and optimized KV caching. Contributions include: ranking 6th overall on Chatbot Arena, and 2nd–4th on Chinese, mathematical, coding, and hard-prompt benchmarks; substantial reduction in inference cost; empirical identification of misalignment between static benchmarks and human preferences—spurring evaluation paradigm innovation; and competitive performance against state-of-the-art closed-source models across major academic benchmarks.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Computer Vision: Large Vision Models

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

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.

Problem

• Super Language Model
• Efficiency and Cost-effectiveness
• Chinese and Multidisciplinary Proficiency

Innovation

• Super Language Model
• Advanced Structural Optimization
• Cost-Efficient Training