The Mixture of Experts (MoE) framework has become a popular architecture for large language models due to its superior performance over dense models. However, training MoEs from scratch in a large-scale regime is prohibitively expensive.

The Mixture-of-Experts (MoE) paradigm has emerged as a powerful approach for scaling transformers with improved resource utilization. However, efficiently fine-tuning MoE models remains largely underexplored. Inspired by recent works on Parameter-Efficient Fine-Tuning (PEFT), we present a unified framework for integrating PEFT modules directly into the MoE mechanism. Aligning with the core principles and architecture of MoE, our framework encompasses a set of design dimensions including various functional and composition strategies. By combining design choices within our framework, we introduce Parameter-Efficient Routed Fine-Tuning (PERFT) as a flexible and scalable family of PEFT strategies tailored for MoE models. Extensive experiments on adapting OLMoE-1B-7B and Mixtral-8$\times$7B for commonsense and arithmetic reasoning tasks demonstrate the effectiveness, scalability, and intriguing dynamics of PERFT. Additionally, we provide empirical findings for each specific design choice to facilitate better application of MoE and PEFT.

In this work, we aim to simultaneously enhance the effectiveness and efficiency of Mixture-of-Experts (MoE) methods. To achieve this, we propose MoE++, a general and heterogeneous MoE framework that integrates both Feed-Forward Network (FFN) and zero-computation experts. Specifically, we introduce three types of zero-computation experts: the zero expert, copy expert, and constant expert, which correspond to discard, skip, and replace operations, respectively. This design offers three key advantages: (i) Low Computing Overhead: Unlike the uniform mixing mechanism for all tokens within vanilla MoE, MoE++ allows each token to engage with a dynamic number of FFNs, be adjusted by constant vectors, or even skip the MoE layer entirely. Moreover, we leverage gating residuals, enabling each token to consider the pathway taken in the previous layer when selecting the appropriate experts. Extensive experimental results demonstrate that MoE++ achieves better performance while delivering 1.1 2.1 expert forward throughput Large Language Models (LLMs) (Brown et al., 2020; OpenAI, 2022; Ouyang et al., 2022; Chowdhery et al., 2023; Achiam et al., 2023) have achieved substantial advancements, primarily attributed to the expansion of training data and a significant increase in model parameters. Therefore, the Mixtureof-Experts (MoE) architecture (Jacobs et al., 1991; Zhou et al., 2022; Roller et al., 2021), which allows for parameter scaling while keeping computational costs manageable, has become a preferred solution. The recent incorporation of MoE architectures into Transformers (Vaswani et al., 2017) has enabled the effective scaling of language models to impressive sizes, resulting in exceptional performance (Team, 2024; Dai et al., 2024; Jiang et al., 2024; Shen et al., 2024; Wei et al., 2024). These achievements underscore the significant potential and promise of MoE language models. Most existing Mixture-of-Experts (MoE) methods (Du et al., 2022; Fedus et al., 2022; Lewis et al., 2021; Rajbhandari et al., 2022) typically activate a fixed number of Feed-Forward Networks (FFNs) for all tokens. In many works (Lepikhin et al., 2021; Xue et al., 2024), each token selects the top We define expert throughput as the throughput of FFN experts and zero-computation experts (if present).

The Mixture of Experts (MoE) framework has become a popular architecture for large language models due to its superior performance over dense models. However, training MoEs from scratch in a large-scale regime is prohibitively expensive. Existing methods mitigate this by pre-training multiple dense expert models independently and using them to initialize an MoE. This is done by using experts' feed-forward network (FFN) to initialize the MoE's experts while merging other parameters. However, this method limits the reuse of dense model parameters to only the FFN layers, thereby constraining the advantages when "upcycling" these models into MoEs. We propose BAM (Branch-Attend-Mix), a simple yet effective method that addresses this shortcoming. BAM makes full use of specialized dense models by not only using their FFN to initialize the MoE layers but also leveraging experts' attention parameters fully by initializing them into a soft-variant of Mixture of Attention (MoA) layers. We explore two methods for upcycling attention parameters: 1) initializing separate attention experts from dense models including all attention parameters for the best model performance; and 2) sharing key and value parameters across all experts to facilitate for better inference efficiency. To further improve efficiency, we adopt a parallel attention transformer architecture to MoEs, which allows the attention experts and FFN experts to be computed concurrently. Our experiments on seed models ranging from 590 million to 2 billion parameters demonstrate that BAM surpasses baselines in both perplexity and downstream task performance, within the same computational and data constraints.

When GPT-3 was introduced by OpenAI in May 2020 the news spread like wildfire. Not only amongst the AI community but even within the mainstream media there were headlines like "A robot wrote this article" and "Have you read something written by GPT-3?". Before GPT-3, the largest language model was Turing-NLG with 17 billion parameters, released in February 2020. Later that year, OpenAI blew this out the park with 175 billion parameters. Suddenly, there was a language model that could produce content that was often indistinguishable from humans.