复用旧模型检查点，加速大模型微调。
Mashup Learning: Faster Finetuning by Remixing Past Checkpoints
- 从历史训练结果中挑选相关检查点，通过合并生成新初始化权重。
- 在8个基准上平均提升0.5-5个百分点准确率，减少41%-46%训练步数。
- 适合需要快速微调且有历史模型的团队，显著节省时间和资源。
在特定领域数据上微调是提升大语言模型下游任务性能的常用方法。每次训练都会产生新的模型权重，形成大量保存在本地或开源平台的检查点。然而，这些训练产物很少被用于后续实验，尽管它们可能包含对相似任务有益的改进能力。本文提出Mashup Learning，一种简单方法，利用先前训练结果来加速模型对新任务的适应。该方法识别与目标数据集最相关的过往检查点，通过模型合并进行聚合，并将结果作为新训练的优化初始值。在8个标准大模型基准、4种模型和两组源检查点上，Mashup Learning相比从头训练，平均下游准确率提升0.5-5个百分点。同时加速收敛，仅需41%-46%的训练步数即可达到从头训练的准确率，总墙钟时间最多减少37%，包含选择与合并开销。
原文摘要 · Abstract (English)
Finetuning on domain-specific data is a well-established method for enhancing LLM performance on downstream tasks. Training on each dataset produces a new set of model weights, resulting in a multitude of checkpoints saved in-house or on open-source platforms. However, these training artifacts are rarely reused for subsequent experiments despite containing improved model abilities for potentially similar tasks. In this paper, we propose Mashup Learning, a simple method to leverage the outputs of prior training runs to enhance model adaptation to new tasks. Our procedure identifies the most relevant historical checkpoints for a target dataset, aggregates them with model merging, and uses the result as an improved initialization for training. Across 8 standard LLM benchmarks, four models, and two collections of source checkpoints, Mashup Learning consistently improves average downstream accuracy by 0.5-5 percentage points over training from scratch. It also accelerates convergence, requiring 41-46% fewer training steps and up to 37% less total wall-clock time to match from-scratch accuracy, including all selection and merging overhead.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。