LLaMA-MoE: Building Mixture-of-Experts from LLaMA with Continual Pre-training
This work addresses the challenge of efficiently scaling large language models for AI researchers and practitioners, but it is incremental as it adapts an existing model rather than introducing a new paradigm.
The paper tackles the data-hungry and instability problems in training large-scale Mixture-of-Experts (MoE) models by building MoE from the existing LLaMA-2 7B model through expert construction and continual pre-training, resulting in LLaMA-MoE-3.5B models that significantly outperform dense models with similar activation parameters after training on 200B tokens.
Mixture-of-Experts (MoE) has gained increasing popularity as a promising framework for scaling up large language models (LLMs). However, training MoE from scratch in a large-scale setting still suffers from data-hungry and instability problems. Motivated by this limit, we investigate building MoE models from existing dense large language models. Specifically, based on the well-known LLaMA-2 7B model, we obtain an MoE model by: (1) Expert Construction, which partitions the parameters of original Feed-Forward Networks (FFNs) into multiple experts; (2) Continual Pre-training, which further trains the transformed MoE model and additional gate networks. In this paper, we comprehensively explore different methods for expert construction and various data sampling strategies for continual pre-training. After these stages, our LLaMA-MoE models could maintain language abilities and route the input tokens to specific experts with part of the parameters activated. Empirically, by training 200B tokens, LLaMA-MoE-3.5B models significantly outperform dense models that contain similar activation parameters. The source codes and models are available at https://github.com/pjlab-sys4nlp/llama-moe .