Pre-training under infinite compute

• URL: http://arxiv.org/abs/2509.14786v1
• Date: Thu, 18 Sep 2025 09:36:23 GMT
• Title: Pre-training under infinite compute
• Authors: Konwoo Kim, Suhas Kotha, Percy Liang, Tatsunori Hashimoto,
• Abstract summary: We show that data-constrained approaches of increasing epoch count and parameter count eventually overfit.<n>We identify that ensembling independently trained models achieves a significantly lower loss asymptote than the regularized recipe.<n>Our results show that simple algorithmic improvements can enable significantly more data-efficient pre-training in a compute-rich future.
• Score: 87.02472603429936
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Since compute grows much faster than web text available for language model pre-training, we ask how one should approach pre-training under fixed data and no compute constraints. We first show that existing data-constrained approaches of increasing epoch count and parameter count eventually overfit, and we significantly improve upon such recipes by properly tuning regularization, finding that the optimal weight decay is $30\times$ larger than standard practice. Since our regularized recipe monotonically decreases loss following a simple power law in parameter count, we estimate its best possible performance via the asymptote of its scaling law rather than the performance at a fixed compute budget. We then identify that ensembling independently trained models achieves a significantly lower loss asymptote than the regularized recipe. Our best intervention combining epoching, regularization, parameter scaling, and ensemble scaling achieves an asymptote at 200M tokens using $5.17\times$ less data than our baseline, and our data scaling laws predict that this improvement persists at higher token budgets. We find that our data efficiency gains can be realized at much smaller parameter counts as we can distill an ensemble into a student model that is 8$\times$ smaller and retains $83\%$ of the ensembling benefit. Finally, our interventions designed for validation loss generalize to downstream benchmarks, achieving a $9\%$ improvement for pre-training evals and a $17.5\times$ data efficiency improvement over continued pre-training on math mid-training data. Our results show that simple algorithmic improvements can enable significantly more data-efficient pre-training in a compute-rich future.

Related papers

• Less is More: Convergence Benefits of Fewer Data Weight Updates over Longer Horizon [42.1998022417145]
  We analyze the convergence behavior of data mixing with a finite number of inner steps $T$.<n>We show that the optimal $T$ scales as $(log N)$ (resp., $((N log N)1/2)$) for the data mixing problem with access to full.
  arXiv  Detail & Related papers  (2026-02-23T04:50:13Z)
• NeuCLIP: Efficient Large-Scale CLIP Training with Neural Normalizer Optimization [42.298647858844895]
  Accurately estimating the normalization term in the contrastive loss is a central challenge for Contrastive Language-Image Pre-training models.<n>We propose NeuCLIP, a novel and elegant optimization framework based on two key ideas.<n>Experiments on large-scale CLIP training, spanning datasets from millions to billions of samples, demonstrate that NeuCLIP outperforms previous methods.
  arXiv  Detail & Related papers  (2025-11-11T16:27:51Z)
• 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.<n>We find that initiating pruning at 25% of total training compute and concluding at 75% achieves near-optimal final evaluation loss.<n>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)
• Scaling Data-Constrained Language Models [133.2083255645999]
  We investigate scaling language models in data-constrained regimes.<n>We find that with constrained data for a fixed compute budget, training with up to 4 epochs of repeated data yields negligible changes to loss compared to having unique data.<n>We propose and empirically validate a scaling law for compute optimality that accounts for the decreasing value of repeated tokens and excess parameters.
  arXiv  Detail & Related papers  (2023-05-25T17:18:55Z)
• Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learning [81.3514358542452]
  Few-shot in-context learning (ICL) incurs substantial computational, memory, and storage costs because it involves processing all of the training examples every time a prediction is made. parameter-efficient fine-tuning offers an alternative paradigm where a small set of parameters are trained to enable a model to perform the new task. In this paper, we rigorously compare few-shot ICL and parameter-efficient fine-tuning and demonstrate that the latter offers better accuracy as well as dramatically lower computational costs.
  arXiv  Detail & Related papers  (2022-05-11T17:10:41Z)
• Scaling Laws for Transfer [0.5432984841650929]
  We study scaling laws for transfer learning between distributions in an unsupervised, fine-tuning setting. We find that the effective data transferred is described well in the low data regime by a power-law of parameter count and fine-tuning dataset size.
  arXiv  Detail & Related papers  (2021-02-02T04:07:38Z)
• Predicting Training Time Without Training [120.92623395389255]
  We tackle the problem of predicting the number of optimization steps that a pre-trained deep network needs to converge to a given value of the loss function. We leverage the fact that the training dynamics of a deep network during fine-tuning are well approximated by those of a linearized model. We are able to predict the time it takes to fine-tune a model to a given loss without having to perform any training.
  arXiv  Detail & Related papers  (2020-08-28T04:29:54Z)
• 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)

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.