Ruyi2 Technical Report

TL;DR

Ruyi2 is an adaptive large language model that leverages familial parameter sharing and 3D parallel training to significantly improve efficiency and scalability while maintaining high performance.

Contribution

It introduces a stable familial model architecture with 3D parallel training, enabling efficient variable-depth computation and scalable deployment of large language models.

Findings

Achieves 2-3x speedup over Ruyi model

Performs comparably to Qwen3 models of similar size

Establishes a 'Train Once, Deploy Many' paradigm

Abstract

Large Language Models (LLMs) face significant challenges regarding deployment costs and latency, necessitating adaptive computing strategies. Building upon the AI Flow framework, we introduce Ruyi2 as an evolution of our adaptive model series designed for efficient variable-depth computation. While early-exit architectures offer a viable efficiency-performance balance, the Ruyi model and existing methods often struggle with optimization complexity and compatibility with large-scale distributed training. To bridge this gap, Ruyi2 introduces a stable "Familial Model" based on Megatron-LM. By using 3D parallel training, it achieves a 2-3 times speedup over Ruyi, while performing comparably to same-sized Qwen3 models. These results confirm that family-based parameter sharing is a highly effective strategy, establishing a new "Train Once, Deploy Many" paradigm and providing a key reference for balancing architectural efficiency with high-performance capabilities.