How Much is Enough? The Diminishing Returns of Tokenization Training Data

• URL: http://arxiv.org/abs/2502.20273v4
• Date: Mon, 16 Jun 2025 16:51:10 GMT
• Title: How Much is Enough? The Diminishing Returns of Tokenization Training Data
• Abstract summary: We train BPE, UnigramLM, and WordPiece tokenizers across various vocabulary sizes using English training data ranging from 1GB to 900GB.<n>Our findings reveal diminishing returns as training data size increases beyond roughly 150GB.<n>For Russian text, we observe diminishing returns after training a tokenizer from 200GB of data, which is approximately 33% more than when training on English.
• Score: 4.300681074103876
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Tokenization, a crucial initial step in natural language processing, is governed by several key parameters, such as the tokenization algorithm, vocabulary size, pre-tokenization strategy, inference strategy, and training data corpus. This paper investigates the impact of an often-overlooked hyperparameter, tokenizer training data size. We train BPE, UnigramLM, and WordPiece tokenizers across various vocabulary sizes using English training data ranging from 1GB to 900GB. Our findings reveal diminishing returns as training data size increases beyond roughly 150GB, suggesting a practical limit to the improvements in tokenization quality achievable through additional data. We analyze this phenomenon and attribute the saturation effect to constraints introduced by the pre-tokenization stage. We then demonstrate the extent to which these findings can generalize by experimenting on data in Russian, a language typologically distant from English. For Russian text, we observe diminishing returns after training a tokenizer from 200GB of data, which is approximately 33% more than when training on English. These results provide valuable insights for optimizing the tokenization process by reducing the compute required for training on large corpora and suggest promising directions for future research in tokenization algorithms.

Related papers

• Compute Optimal Tokenization [79.3815358070537]
  We investigate how the information granularity of tokens, controlled by the compression rate, affects scaling trends.<n>We train 988 latent tokenized models (BLT) ranging from 50M to 7B parameters that enable setting the desired compression rate.<n>Our experiments reveal that in compute-optimal configurations, model parameter counts scale proportionally to data size measured in bytes, not in tokens as commonly perceived.
  arXiv  Detail & Related papers  (2026-05-02T01:53:22Z)
• Simultaneous Speech-to-Speech Translation Without Aligned Data [52.467808474293605]
  Simultaneous speech translation requires translating source speech into a target language in real-time.<n>We propose Hibiki-Zero, which eliminates the need for word-level alignments entirely.<n>Hibiki-Zero achieves state-of-the-art performance in translation accuracy, latency, voice transfer, and naturalness across five X-to-English tasks.
  arXiv  Detail & Related papers  (2026-02-11T17:41:01Z)
• AdaptBPE: From General Purpose to Specialized Tokenizers [18.70903226766322]
  We propose a post-training adaptation strategy that selectively replaces low-utility tokens with more relevant ones based on their frequency in an adaptation corpus.<n>Our algorithm identifies the token inventory that most effectively encodes the adaptation corpus for a given target vocabulary size.<n>This method serves as a lightweight adaptation mechanism, akin to a vocabulary fine-tuning process, enabling optimized tokenization for specific domains or tasks.
  arXiv  Detail & Related papers  (2026-01-29T12:59:40Z)
• Explaining and Mitigating Crosslingual Tokenizer Inequities [18.823984182115712]
  Having high token premiums leads to less throughput during training and increases costs at inference.<n>We train a suite of approximately 7,000 comparable monolingual tokenizers for 97 languages.<n>We find that similarity between training and test data does not impact token premiums, but vocabulary size and pre-tokenization do.
  arXiv  Detail & Related papers  (2025-10-24T17:36:03Z)
• Comparative analysis of subword tokenization approaches for Indian languages [5.012314384895538]
  Tokenization is the act of breaking down text into smaller parts, or tokens, that are easier for machines to process.<n>Subword tokenization enhances this process by breaking down words into smaller subword units.<n>It is useful in capturing the intricate structure of words in Indian languages (ILs), such as prefixes, suffixes, and other morphological variations.<n>This paper examines how different subword tokenization techniques, such as SentencePiece, Byte Pair, and WordPiece Tokenization, affect ILs.
  arXiv  Detail & Related papers  (2025-05-22T16:24:37Z)
• Scaling Laws for Data-Efficient Visual Transfer Learning [14.114908296325277]
  This paper establishes the first practical framework for data-efficient scaling laws in visual transfer learning. We propose the distillation boundary theory, revealing a critical turning point in distillation efficiency. This work redefines scaling laws for data-limited regimes, bridging the knowledge gap between large-scale pretraining and practical downstream adaptation.
  arXiv  Detail & Related papers  (2025-04-17T07:01:01Z)
• Reasoning to Learn from Latent Thoughts [45.59740535714148]
  We show that explicitly modeling and inferring the latent thoughts that underlie the text generation process can significantly improve pretraining data efficiency. We show that a 1B LM can bootstrap its performance across at least three iterations and significantly outperform baselines trained on raw data. The gains from inference scaling and EM iterations suggest new opportunities for scaling data-constrained pretraining.
  arXiv  Detail & Related papers  (2025-03-24T16:41:23Z)
• Three Things to Know about Deep Metric Learning [34.16300515811057]
  This paper addresses supervised deep metric learning for open-set image retrieval.<n>It focuses on three key aspects: the loss function, mixup regularization, and model initialization.<n>Through a systematic study of these components, we demonstrate that their synergy enables large models to nearly solve popular benchmarks.
  arXiv  Detail & Related papers  (2024-12-17T00:49:12Z)
• Capturing the Temporal Dependence of Training Data Influence [100.91355498124527]
  We formalize the concept of trajectory-specific leave-one-out influence, which quantifies the impact of removing a data point during training.<n>We propose data value embedding, a novel technique enabling efficient approximation of trajectory-specific LOO.<n>As data value embedding captures training data ordering, it offers valuable insights into model training dynamics.
  arXiv  Detail & Related papers  (2024-12-12T18:28:55Z)
• Batching BPE Tokenization Merges [55.2480439325792]
  BatchBPE is an open-source pure Python implementation of the Byte Pair algorithm. It is used to train a high quality tokenizer on a basic laptop.
  arXiv  Detail & Related papers  (2024-08-05T09:37:21Z)
• Data Mixture Inference: What do BPE Tokenizers Reveal about their Training Data? [112.0422370149713]
  We tackle a task which we call data mixture inference, which aims to uncover the distributional make-up of training data.<n>We introduce a novel attack based on a previously overlooked source of information: byte-pair encoding (BPE) tokenizers.<n>We show that our attack recovers mixture ratios with high precision for tokenizers trained on known mixtures of natural languages, programming languages, and data sources.
  arXiv  Detail & Related papers  (2024-07-23T16:13:22Z)
• Unpacking Tokenization: Evaluating Text Compression and its Correlation with Model Performance [34.641079276516926]
  We argue for the theoretical importance of compression, that can be viewed as 0-gram language modeling. We demonstrate the empirical importance of compression for downstream success of pre-trained language models. We show that there is a correlation between tokenizers' compression and models' downstream performance.
  arXiv  Detail & Related papers  (2024-03-10T17:02:53Z)
• Tokenization Is More Than Compression [14.939912120571728]
  Existing tokenization approaches like Byte-Pair. (BPE) originate from the field of data compression. We introduce PathPiece, a new tokenizer that segments a document's text into the minimum number of tokens for a given vocabulary.
  arXiv  Detail & Related papers  (2024-02-28T14:52:15Z)
• Disparate Impact on Group Accuracy of Linearization for Private Inference [48.27026603581436]
  We show that reducing the number of ReLU activations disproportionately decreases the accuracy for minority groups compared to majority groups. We also show how a simple procedure altering the fine-tuning step for linearized models can serve as an effective mitigation strategy.
  arXiv  Detail & Related papers  (2024-02-06T01:56:29Z)
• Online Importance Sampling for Stochastic Gradient Optimization [33.42221341526944]
  We propose a practical algorithm that efficiently computes data importance on-the-fly during training.<n>We also introduce a novel metric based on the derivative of the loss w.r.t. the network output, designed for mini-batch importance sampling.
  arXiv  Detail & Related papers  (2023-11-24T13:21:35Z)
• Small batch deep reinforcement learning [31.69289254478042]
  In value-based deep reinforcement learning, the batch size parameter specifies how many transitions to sample for each gradient update. In this work we present a broad empirical study that suggests em reducing the batch size can result in a number of significant performance gains.
  arXiv  Detail & Related papers  (2023-10-05T20:31:37Z)
• 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)
• To Repeat or Not To Repeat: Insights from Scaling LLM under Token-Crisis [50.31589712761807]
  Large language models (LLMs) are notoriously token-hungry during pre-training, and high-quality text data on the web is approaching its scaling limit for LLMs. We investigate the consequences of repeating pre-training data, revealing that the model is susceptible to overfitting. Second, we examine the key factors contributing to multi-epoch degradation, finding that significant factors include dataset size, model parameters, and training objectives.
  arXiv  Detail & Related papers  (2023-05-22T17:02:15Z)
• An Information Extraction Study: Take In Mind the Tokenization! [18.20319269401045]
  We study the impact of tokenization when extracting information from documents. We present a comparative study and analysis of subword-based and character-based models. The main outcome is twofold: tokenization patterns can introduce inductive bias that results in state-of-the-art performance.
  arXiv  Detail & Related papers  (2023-03-27T11:08:35Z)
• Time Series Contrastive Learning with Information-Aware Augmentations [57.45139904366001]
  A key component of contrastive learning is to select appropriate augmentations imposing some priors to construct feasible positive samples. How to find the desired augmentations of time series data that are meaningful for given contrastive learning tasks and datasets remains an open question. We propose a new contrastive learning approach with information-aware augmentations, InfoTS, that adaptively selects optimal augmentations for time series representation learning.
  arXiv  Detail & Related papers  (2023-03-21T15:02:50Z)
• Dissecting Continual Learning a Structural and Data Analysis [0.0]
  Continual Learning is a field dedicated to devise algorithms able to achieve lifelong learning. Deep learning methods can attain impressive results when the data modeled does not undergo a considerable distributional shift in subsequent learning sessions. When we expose such systems to this incremental setting, performance drop very quickly.
  arXiv  Detail & Related papers  (2023-01-03T10:37:11Z)
• Memory Augmented Lookup Dictionary based Language Modeling for Automatic Speech Recognition [20.926163659469587]
  We propose a new memory augmented lookup dictionary based Transformer architecture for LM. The newly introduced lookup dictionary incorporates rich contextual information in training set, which is vital to correctly predict long-tail tokens. Our proposed method is proved to outperform the baseline Transformer LM by a great margin on both word/character error rate and tail tokens error rate.
  arXiv  Detail & Related papers  (2022-12-30T22:26:57Z)
• Automatic Data Augmentation via Invariance-Constrained Learning [94.27081585149836]
  Underlying data structures are often exploited to improve the solution of learning tasks. Data augmentation induces these symmetries during training by applying multiple transformations to the input data. This work tackles these issues by automatically adapting the data augmentation while solving the learning task.
  arXiv  Detail & Related papers  (2022-09-29T18:11:01Z)
• How Robust is Neural Machine Translation to Language Imbalance in Multilingual Tokenizer Training? [86.48323488619629]
  We analyze how translation performance changes as the data ratios among languages vary in the tokenizer training corpus. We find that while relatively better performance is often observed when languages are more equally sampled, the downstream performance is more robust to language imbalance than we usually expected.
  arXiv  Detail & Related papers  (2022-04-29T17:50:36Z)
• 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)
• Improving Classifier Training Efficiency for Automatic Cyberbullying Detection with Feature Density [58.64907136562178]
  We study the effectiveness of Feature Density (FD) using different linguistically-backed feature preprocessing methods. We hypothesise that estimating dataset complexity allows for the reduction of the number of required experiments. The difference in linguistic complexity of datasets allows us to additionally discuss the efficacy of linguistically-backed word preprocessing.
  arXiv  Detail & Related papers  (2021-11-02T15:48:28Z)
• Token-wise Curriculum Learning for Neural Machine Translation [94.93133801641707]
  Existing curriculum learning approaches to Neural Machine Translation (NMT) require sufficient sampling amounts of "easy" samples from training data at the early training stage. We propose a novel token-wise curriculum learning approach that creates sufficient amounts of easy samples. Our approach can consistently outperform baselines on 5 language pairs, especially for low-resource languages.
  arXiv  Detail & Related papers  (2021-03-20T03:57:59Z)
• Omni-supervised Facial Expression Recognition via Distilled Data [120.11782405714234]
  We propose omni-supervised learning to exploit reliable samples in a large amount of unlabeled data for network training. We experimentally verify that the new dataset can significantly improve the ability of the learned FER model. To tackle this, we propose to apply a dataset distillation strategy to compress the created dataset into several informative class-wise images.
  arXiv  Detail & Related papers  (2020-05-18T09:36:51Z)
• Machine Translation Pre-training for Data-to-Text Generation -- A Case Study in Czech [5.609443065827995]
  We study the effectiveness of machine translation based pre-training for data-to-text generation in non-English languages. We find that pre-training lets us train end-to-end models with significantly improved performance.
  arXiv  Detail & Related papers  (2020-04-05T02:47:16Z)

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.