Scaling Laws for Upcycling Mixture-of-Experts Language Models

• URL: http://arxiv.org/abs/2502.03009v1
• Date: Wed, 05 Feb 2025 09:11:13 GMT
• Title: Scaling Laws for Upcycling Mixture-of-Experts Language Models
• Authors: Seng Pei Liew, Takuya Kato, Sho Takase,
• Abstract summary: Pretraining large language models (LLMs) is resource-intensive, often requiring months of training time even with high-end GPU clusters. There are two approaches of mitigating such computational demands: reusing smaller models to train larger ones (upcycling) and training computationally efficient models like mixture-of-experts (MoE)
• Score: 17.796361238003403
• License:
• Abstract: Pretraining large language models (LLMs) is resource-intensive, often requiring months of training time even with high-end GPU clusters. There are two approaches of mitigating such computational demands: reusing smaller models to train larger ones (upcycling), and training computationally efficient models like mixture-of-experts (MoE). In this paper, we study the upcycling of LLMs to MoE models, of which the scaling behavior remains underexplored. Through extensive experiments, we identify empirical scaling laws that describe how performance depends on dataset size and model configuration. Particularly, we show that, while scaling these factors improves performance, there is a novel interaction term between the dense and upcycled training dataset that limits the efficiency of upcycling at large computational budgets. Based on these findings, we provide guidance to scale upcycling, and establish conditions under which upcycling outperforms from-scratch trainings within budget constraints.

Related papers

• DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs [70.91804882618243]
  This paper proposes DSMoE, a novel approach that achieves sparsification by partitioning pre-trained FFN layers into computational blocks. We implement adaptive expert routing using sigmoid activation and straight-through estimators, enabling tokens to flexibly access different aspects of model knowledge. Experiments on LLaMA models demonstrate that under equivalent computational constraints, DSMoE achieves superior performance compared to existing pruning and MoE approaches.
  arXiv  Detail & Related papers  (2025-02-18T02:37:26Z)
• Building on Efficient Foundations: Effectively Training LLMs with Structured Feedforward Layers [16.253898272659242]
  State-of-the-art results in large language models (LLMs) often rely on scale, which becomes computationally expensive. Our study focuses on transformer-based LLMs, specifically targeting the computationally intensive feedforward networks (FFNs) We show that wide and structured networks can utilize training FLOPs more efficiently, with fewer parameters and lower loss than dense models at their optimal trade-off.
  arXiv  Detail & Related papers  (2024-06-24T08:43:21Z)
• More Compute Is What You Need [3.184416958830696]
  We propose a new scaling law that suggests model performance depends mostly on the amount of compute spent for transformer-based models. We predict that (a) for inference efficiency, training should prioritize smaller model sizes and larger training datasets, and (b) assuming the exhaustion of available web datasets, scaling the model size might be the only way to further improve model performance.
  arXiv  Detail & Related papers  (2024-04-30T12:05:48Z)
• Scaling Laws For Dense Retrieval [22.76001461620846]
  We investigate whether the performance of dense retrieval models follows the scaling law as other neural models. Results indicate that, under our settings, the performance of dense retrieval models follows a precise power-law scaling related to the model size and the number of annotations.
  arXiv  Detail & Related papers  (2024-03-27T15:27:36Z)
• Mixtures of Experts Unlock Parameter Scaling for Deep RL [54.26191237981469]
  In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules into value-based networks results in more parameter-scalable models. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.
  arXiv  Detail & Related papers  (2024-02-13T17:18:56Z)
• A Dynamical Model of Neural Scaling Laws [79.59705237659547]
  We analyze a random feature model trained with gradient descent as a solvable model of network training and generalization. Our theory shows how the gap between training and test loss can gradually build up over time due to repeated reuse of data.
  arXiv  Detail & Related papers  (2024-02-02T01:41:38Z)
• The LLM Surgeon [33.90611088414982]
  We explore data-driven compression of existing pretrained models as an alternative to training smaller models from scratch. We provide a general framework for unstructured, semi-structured and structured pruning and improve upon weight updates to capture more correlations between weights. Our method can prune rows and columns from a range of OPT models and Llamav2-7B by 20%-30%, with a negligible loss in performance.
  arXiv  Detail & Related papers  (2023-12-28T18:59:09Z)
• The Languini Kitchen: Enabling Language Modelling Research at Different Scales of Compute [66.84421705029624]
  We introduce an experimental protocol that enables model comparisons based on equivalent compute, measured in accelerator hours. We pre-process an existing large, diverse, and high-quality dataset of books that surpasses existing academic benchmarks in quality, diversity, and document length. This work also provides two baseline models: a feed-forward model derived from the GPT-2 architecture and a recurrent model in the form of a novel LSTM with ten-fold throughput.
  arXiv  Detail & Related papers  (2023-09-20T10:31:17Z)
• Scaling Vision-Language Models with Sparse Mixture of Experts [128.0882767889029]
  We show that mixture-of-experts (MoE) techniques can achieve state-of-the-art performance on a range of benchmarks over dense models of equivalent computational cost. Our research offers valuable insights into stabilizing the training of MoE models, understanding the impact of MoE on model interpretability, and balancing the trade-offs between compute performance when scaling vision-language models.
  arXiv  Detail & Related papers  (2023-03-13T16:00:31Z)
• Convolutional Tensor-Train LSTM for Spatio-temporal Learning [116.24172387469994]
  We propose a higher-order LSTM model that can efficiently learn long-term correlations in the video sequence. This is accomplished through a novel tensor train module that performs prediction by combining convolutional features across time. Our results achieve state-of-the-art performance-art in a wide range of applications and datasets.
  arXiv  Detail & Related papers  (2020-02-21T05:00:01Z)
• Scaling Laws for Neural Language Models [14.472857826717613]
  We study scaling laws for language model performance on the cross-entropy loss. The loss scales as a power-law with model size, dataset size, and the amount of compute used for training.
  arXiv  Detail & Related papers  (2020-01-23T03:59:20Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.