AutoScale: Scale-Aware Data Mixing for Pre-Training LLMs

• URL: http://arxiv.org/abs/2407.20177v4
• Date: Sun, 06 Apr 2025 03:22:39 GMT
• Title: AutoScale: Scale-Aware Data Mixing for Pre-Training LLMs
• Authors: Feiyang Kang, Yifan Sun, Bingbing Wen, Si Chen, Dawn Song, Rafid Mahmood, Ruoxi Jia,
• Abstract summary: We show that data mixtures that perform well at smaller scales may not retain their advantage at larger scales.<n>We propose AutoScale, a two-stage, scale-aware data composition framework.
• Score: 61.13296177652599
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Domain reweighting is an emerging research area aimed at adjusting the relative weights of different data sources to improve the effectiveness and efficiency of LLM pre-training. We show that data mixtures that perform well at smaller scales may not retain their advantage at larger scales, challenging the existing practice of determining competitive mixtures in small-scale experiments and directly applying them at much larger scales. To address this, we propose AutoScale, a two-stage, scale-aware data composition framework. First, AutoScale fits a parametric model that predicts the model's loss under different data compositions, then uses it to find an approximate best allocation at smaller, more manageable budgets. Next, leveraging a novel theoretical analysis of how optimal compositions evolve with scale, AutoScale extrapolates that composition to larger budgets without further retraining. Empirically, AutoScale accelerates convergence and improves downstream performance. For instance, when pre-training GPT-2 Large, it achieves a 28% faster perplexity reduction than baselines and up to a 38% speed-up over unweighted training, while yielding best-average results on various downstream tasks. Overall, our findings illustrate how domain importance shifts with training scale, underscoring the need for scale-dependent data curation in LLM training. Our code is open-sourced.

Related papers

• Optimizing ML Training with Metagradient Descent [69.89631748402377]
  We introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a "smooth model training" framework that enables effective optimization using metagradients.
  arXiv  Detail & Related papers  (2025-03-17T22:18:24Z)
• Communication-Efficient Language Model Training Scales Reliably and Robustly: Scaling Laws for DiLoCo [22.7130140114906]
  We study the scaling law behavior of DiLoCo when training LLMs under a fixed compute budget. We find that DiLoCo scales both predictably and robustly with model size. When well-tuned, DiLoCo scales better than data-parallel training with model size, and can outperform data-parallel training even at small model sizes.
  arXiv  Detail & Related papers  (2025-03-12T20:04:38Z)
• LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws [21.053622641336744]
  Loss-to-loss scaling laws relate losses across pretraining datasets and downstream tasks. Our experiments reveal that the pretraining data and tokenizer determine the scaling trend.
  arXiv  Detail & Related papers  (2025-02-17T18:45:25Z)
• The Journey Matters: Average Parameter Count over Pre-training Unifies Sparse and Dense Scaling Laws [51.608402959163925]
  We present the first systematic exploration of optimal sparse pre-training configurations for large language models. We find that initiating pruning at 25% of total training compute and concluding at 75% achieves near-optimal final evaluation loss. We propose a new scaling law that modifies the Chinchilla scaling law to use the average parameter count over pre-training.
  arXiv  Detail & Related papers  (2025-01-21T20:23:22Z)
• Optimizing Pretraining Data Mixtures with LLM-Estimated Utility [52.08428597962423]
  Large Language Models improve with increasing amounts of high-quality training data. We find token-counts outperform manual and learned mixes, indicating that simple approaches for dataset size and diversity are surprisingly effective. We propose two complementary approaches: UtiliMax, which extends token-based $200s by incorporating utility estimates from reduced-scale ablations, achieving up to a 10.6x speedup over manual baselines; and Model Estimated Data Utility (MEDU), which leverages LLMs to estimate data utility from small samples, matching ablation-based performance while reducing computational requirements by $simx.
  arXiv  Detail & Related papers  (2025-01-20T21:10:22Z)
• Optimizing LLMs with Direct Preferences: A Data Efficiency Perspective [4.548047308860141]
  This study investigates the impact of different type of preference data on model performance. It aims to reduce their dependency on extensive amounts of preference data, which is expensive to collect.
  arXiv  Detail & Related papers  (2024-10-22T00:11:41Z)
• Optimizing importance weighting in the presence of sub-population shifts [0.0]
  A distribution shift between the training and test data can severely harm performance of machine learning models. We argue that existing weightings for determining the weights are suboptimal, as they neglect the increase of the variance of the estimated model due to the finite sample size of the training data. We propose a bi-level optimization procedure in which the weights and model parameters are optimized simultaneously.
  arXiv  Detail & Related papers  (2024-10-18T09:21:10Z)
• Fisher Information-based Efficient Curriculum Federated Learning with Large Language Models [43.26028399395612]
  We propose a Fisher Information-based Efficient Curriculum Federated Learning framework (FibecFed) with two novel methods. First, we propose a fisher information-based method to adaptively sample data within each device to improve the effectiveness of the FL fine-tuning process. Second, we dynamically select the proper layers for global aggregation and sparse parameters for local update with LoRA.
  arXiv  Detail & Related papers  (2024-09-30T18:12:18Z)
• An Emulator for Fine-Tuning Large Language Models using Small Language Models [91.02498576056057]
  We introduce emulated fine-tuning (EFT), a principled and practical method for sampling from a distribution that approximates the result of pre-training and fine-tuning at different scales. We show that EFT enables test-time adjustment of competing behavioral traits like helpfulness and harmlessness without additional training. Finally, a special case of emulated fine-tuning, which we call LM up-scaling, avoids resource-intensive fine-tuning of large pre-trained models by ensembling them with small fine-tuned models.
  arXiv  Detail & Related papers  (2023-10-19T17:57:16Z)
• Scaling Laws for Sparsely-Connected Foundation Models [70.41266138010657]
  We explore the impact of parameter sparsity on the scaling behavior of Transformers trained on massive datasets. We identify the first scaling law describing the relationship between weight sparsity, number of non-zero parameters, and amount of training data.
  arXiv  Detail & Related papers  (2023-09-15T16:29:27Z)
• D4: Improving LLM Pretraining via Document De-Duplication and Diversification [38.84592304799403]
  We show that careful data selection via pre-trained model embeddings can speed up training. We also show that repeating data intelligently consistently outperforms baseline training.
  arXiv  Detail & Related papers  (2023-08-23T17:58:14Z)
• SPDF: Sparse Pre-training and Dense Fine-tuning for Large Language Models [4.114555639014612]
  We show the benefits of using unstructured weight sparsity to train only a subset of weights during pre-training. We demonstrate that we can induce up to 75% sparsity into a 1.3B parameter GPT-3 XL model resulting in a 2.5x reduction in pre-training FLOPs.
  arXiv  Detail & Related papers  (2023-03-18T17:56:01Z)
• Improved Fine-tuning by Leveraging Pre-training Data: Theory and Practice [52.11183787786718]
  Fine-tuning a pre-trained model on the target data is widely used in many deep learning applications. Recent studies have empirically shown that training from scratch has the final performance that is no worse than this pre-training strategy. We propose a novel selection strategy to select a subset from pre-training data to help improve the generalization on the target task.
  arXiv  Detail & Related papers  (2021-11-24T06:18:32Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)
• Dynamic Scale Training for Object Detection [111.33112051962514]
  We propose a Dynamic Scale Training paradigm (abbreviated as DST) to mitigate scale variation challenge in object detection. Experimental results demonstrate the efficacy of our proposed DST towards scale variation handling. It does not introduce inference overhead and could serve as a free lunch for general detection configurations.
  arXiv  Detail & Related papers  (2020-04-26T16:48:17Z)

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.