More Like this

1. Data mixture inference attack: BPE tokenizers reveal training data compositions

   Hayase, Jonathan; Liu, Alisa; Choi, Yejin; Oh, Sewoong; Smith, Noah A (December 2024, Advances in Neural Information Processing Systems)

   The pretraining data of today's strongest language models remains opaque, even when their parameters are open-sourced. In particular, little is known about the proportions of different domains, languages, or code represented in the data. While a long line of membership inference attacks aim to identify training examples on an instance level, they do not extend easily to global statistics about the corpus. In this work, we tackle a task which we call data mixture inference, which aims to uncover the distributional make-up of the pretraining data. We introduce a novel attack based on a previously overlooked source of information—byte-pair encoding (BPE) tokenizers, used by the vast majority of modern language models. Our key insight is that the ordered vocabulary learned by a BPE tokenizer naturally reveals information about the token frequencies in its training data: the first token is the most common byte pair, the second is the most common pair after merging the first token, and so on. Given a tokenizer's merge list along with data samples for each category of interest (eg, different natural languages), we formulate a linear program that solves for the relative proportion of each category in the tokenizer's training set. Importantly, to the extent to which tokenizer training data is representative of the pretraining data, we indirectly learn about the pretraining data. In controlled experiments, we show that our attack can recover mixture ratios with high precision for tokenizers trained on known mixtures of natural languages, programming languages, and data sources. We then apply our approach to off-the-shelf tokenizers released alongside recent LMs. We confirm much publicly disclosed information about these models, and also make several new inferences: GPT-4o is much more multilingual than its predecessors, training on 10x more non-English data than GPT-3.5, Llama 3 and Claude are trained on predominantly code, and many recent models are trained on 7-16% books. We hope our work sheds light on current design practices for pretraining data, and inspires continued research into data mixture inference for LMs. 

   more » « less
2. Sampling from Your Language Model One Byte at a Time

   Hayase, Jonathan; Liu, Alisa; Smith, Noah A; Oh, Sewoong (July 2025, cs.CL)

   Tokenization is used almost universally by modern language models, enabling efficient text representation using multi-byte or multi-character tokens. However, prior work has shown that tokenization can introduce distortion into the model's generations, an issue known as the Prompt Boundary Problem (PBP). For example, users are often advised not to end their prompts with a space because it prevents the model from including the space as part of the next token. While this heuristic is effective in English, the underlying PBP continues to affect languages such as Chinese as well as code generation, where tokens often do not line up with word and syntactic boundaries. In this work, we present an inference-time method to convert any autoregressive LM with a BPE tokenizer into a character-level or byte-level LM. Our method efficiently solves the PBP and is also able to unify the vocabularies of language models with different tokenizers, allowing one to ensemble LMs with different tokenizers at inference time or transfer the post-training from one model to another using proxy-tuning. We demonstrate in experiments that the ensemble and proxy-tuned models outperform their constituents on downstream evals. Code is available at this https URL . 

   more » « less
3. SuperBPE: Space Travel for Language Models

   Liu, A; Hayase, J; Hofmann, V; Oh, S; Smith, N A; Choi, Y (April 2025, https://doi.org/10.48550/arXiv.2503.13423)

   The assumption across nearly all language model (LM) tokenization schemes is that tokens should be subwords, i.e., contained within word boundaries. While providing a seemingly reasonable inductive bias, is this common practice limiting the potential of modern LMs? Whitespace is not a reliable delimiter of meaning, as evidenced by multi-word expressions (e.g., "by the way"), crosslingual variation in the number of words needed to express a concept (e.g., "spacesuit helmet" in German is "raumanzughelm"), and languages that do not use whitespace at all (e.g., Chinese). To explore the potential of tokenization beyond subwords, we introduce a "superword" tokenizer, SuperBPE, which incorporates a simple pretokenization curriculum into the byte-pair encoding (BPE) algorithm to first learn subwords, then superwords that bridge whitespace. This brings dramatic improvements in encoding efficiency: when fixing the vocabulary size to 200k, SuperBPE encodes a fixed piece of text with up to 33% fewer tokens than BPE on average. In experiments, we pretrain 8B transformer LMs from scratch while fixing the model size, vocabulary size, and train compute, varying *only* the algorithm for learning the vocabulary. Our model trained with SuperBPE achieves an average +4.0% absolute improvement over the BPE baseline across 30 downstream tasks (including +8.2% on MMLU), while simultaneously requiring 27% less compute at inference time. In analysis, we find that SuperBPE results in segmentations of text that are more uniform in per-token difficulty. Qualitatively, this may be because SuperBPE tokens often capture common multi-word expressions that function semantically as a single unit. SuperBPE is a straightforward, local modification to tokenization that improves both encoding efficiency and downstream performance, yielding better language models overall. 

   more » « less
4. Multi-Source Cross-Lingual Model Transfer: Learning What to Share

   Chen, Xilun; Hassan Awadallah, Ahmed; Hassan, Hany; Wang, Wei; Cardie, Claire (January 2019, Proceedings of the 57th Conference of the Association for Computational Linguistics (ACL))

   Modern NLP applications have enjoyed a great boost utilizing neural networks models. Such deep neural models, however, are not applicable to most human languages due to the lack of annotated training data for various NLP tasks. Cross-lingual transfer learning (CLTL) is a viable method for building NLP models for a low-resource target language by leveraging labeled data from other (source) languages. In this work, we focus on the multilingual transfer setting where training data in multiple source languages is leveraged to further boost target language performance. Unlike most existing methods that rely only on language-invariant features for CLTL, our approach coherently utilizes both language invariant and language-specific features at instance level. Our model leverages adversarial networks to learn language-invariant features, and mixture-of-experts models to dynamically exploit the similarity between the target language and each individual source language1. This enables our model to learn effectively what to share between various languages in the multilingual setup. Moreover, when coupled with unsupervised multilingual embeddings, our model can operate in a zero-resource setting where neither target language training data nor cross-lingual resources are available. Our model achieves significant performance gains over prior art, as shown in an extensive set of experiments over multiple text classification and sequence tagging. 

   more » « less
5. CLIPTrans: Transferring Visual Knowledge with Pre-trained Models for Multimodal Machine Translation

   https://doi.org/10.1109/ICCV51070.2023.00269  

   Gupta, Devaansh; Kharbanda, Siddhant; Zhou, Jiawei; Li, Wanhua; Pfister, Hanspeter; Wei, Donglai (October 2023, Proceedings)

   There has been a growing interest in developing multimodal machine translation (MMT) systems that enhance neural machine translation (NMT) with visual knowledge. This problem setup involves using images as auxiliary information during training, and more recently, eliminating their use during inference. Towards this end, previous works face a challenge in training powerful MMT models from scratch due to the scarcity of annotated multilingual vision-language data, especially for low-resource languages. Simultaneously, there has been an influx of multilingual pretrained models for NMT and multimodal pre-trained models for vision-language tasks, primarily in English, which have shown exceptional generalisation ability. However, these are not directly applicable to MMT since they do not provide aligned multimodal multilingual features for generative tasks. To alleviate this issue, instead of designing complex modules for MMT, we propose CLIPTrans, which simply adapts the independently pre-trained multimodal M-CLIP and the multilingual mBART. In order to align their embedding spaces, mBART is conditioned on the M-CLIP features by a prefix sequence generated through a lightweight mapping network. We train this in a two-stage pipeline which warms up the model with image captioning before the actual translation task. Through experiments, we demonstrate the merits of this framework and consequently push forward the state-of-the-art across standard benchmarks by an average of +2.67 BLEU. The code can be found at www.github.com/devaansh100/CLIPTrans. 

   more » « less