A Family of LLMs Liberated from Static Vocabularies

• URL: http://arxiv.org/abs/2603.15953v1
• Date: Mon, 16 Mar 2026 22:07:18 GMT
• Title: A Family of LLMs Liberated from Static Vocabularies
• Abstract summary: We present a family of models with up to 70 billion parameters based on the hierarchical autoregressive transformer (HAT) architecture.<n>We show that we can reuse available pre-trained models by converting the Llama 3.1 8B and 70B models into the HAT architecture.<n>We also provide a 7B HAT model, Llama-TFree-HAT-Pretrained, trained entirely from scratch on nearly 4 trillion words.
• Score: 23.053922969985738
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Tokenization is a central component of natural language processing in current large language models (LLMs), enabling models to convert raw text into processable units. Although learned tokenizers are widely adopted, they exhibit notable limitations, including their large, fixed vocabulary sizes and poor adaptability to new domains or languages. We present a family of models with up to 70 billion parameters based on the hierarchical autoregressive transformer (HAT) architecture. In HAT, an encoder transformer aggregates bytes into word embeddings and then feeds them to the backbone, a classical autoregressive transformer. The outputs of the backbone are then cross-attended by the decoder and converted back into bytes. We show that we can reuse available pre-trained models by converting the Llama 3.1 8B and 70B models into the HAT architecture: Llama-3.1-8B-TFree-HAT and Llama-3.1-70B-TFree-HAT are byte-level models whose encoder and decoder are trained from scratch, but where we adapt the pre-trained Llama backbone, i.e., the transformer blocks with the embedding matrix and head removed, to handle word embeddings instead of the original tokens. We also provide a 7B HAT model, Llama-TFree-HAT-Pretrained, trained entirely from scratch on nearly 4 trillion words. The HAT architecture improves text compression by reducing the number of required sequence positions and enhances robustness to intra-word variations, e.g., spelling differences. Through pre-training, as well as subsequent supervised fine-tuning and direct preference optimization in English and German, we show strong proficiency in both languages, improving on the original Llama 3.1 in most benchmarks. We release our models (including 200 pre-training checkpoints) on Hugging Face.

Related papers

• Hierarchical Autoregressive Transformers: Combining Byte- and Word-Level Processing for Robust, Adaptable Language Models [3.382910438968506]
  Tokenization is a fundamental step in natural language processing, breaking text into units that computational models can process.<n>We investigate a hierarchical architecture for autoregressive language modelling that combines character-level and word-level processing.<n>We demonstrate, at scales up to 7 billion parameters, that hierarchical transformers match the downstream task performance of subword-tokenizer-based models.
  arXiv  Detail & Related papers  (2025-01-17T17:51:53Z)
• Byte Latent Transformer: Patches Scale Better Than Tokens [101.10994909832063]
  Byte Latent Transformer (BLT) encodes bytes into dynamically sized patches, which serve as the primary units of computation.<n>For fixed inference costs, BLT shows significantly better scaling than tokenization-based models, by simultaneously growing both patch and model size.
  arXiv  Detail & Related papers  (2024-12-13T05:33:32Z)
• Transformers for Low-Resource Languages:Is F\'eidir Linn! [2.648836772989769]
  In general, neural translation models often under perform on language pairs with insufficient training data. We demonstrate that choosing appropriate parameters leads to considerable performance improvements. A Transformer optimized model demonstrated a BLEU score improvement of 7.8 points when compared with a baseline RNN model.
  arXiv  Detail & Related papers  (2024-03-04T12:29:59Z)
• In-Context Language Learning: Architectures and Algorithms [73.93205821154605]
  We study ICL through the lens of a new family of model problems we term in context language learning (ICLL) We evaluate a diverse set of neural sequence models on regular ICLL tasks.
  arXiv  Detail & Related papers  (2024-01-23T18:59:21Z)
• Improving language models by retrieving from trillions of tokens [50.42630445476544]
  We enhance auto-regressive language models by conditioning on document chunks retrieved from a large corpus. With a $2$ trillion token database, our Retrieval-Enhanced Transformer (RETRO) obtains comparable performance to GPT-3 and Jurassic-1 on the Pile.
  arXiv  Detail & Related papers  (2021-12-08T17:32:34Z)
• Sentence Bottleneck Autoencoders from Transformer Language Models [53.350633961266375]
  We build a sentence-level autoencoder from a pretrained, frozen transformer language model. We adapt the masked language modeling objective as a generative, denoising one, while only training a sentence bottleneck and a single-layer modified transformer decoder. We demonstrate that the sentence representations discovered by our model achieve better quality than previous methods that extract representations from pretrained transformers on text similarity tasks, style transfer, and single-sentence classification tasks in the GLUE benchmark, while using fewer parameters than large pretrained models.
  arXiv  Detail & Related papers  (2021-08-31T19:39:55Z)
• Charformer: Fast Character Transformers via Gradient-based Subword Tokenization [50.16128796194463]
  We propose a new model inductive bias that learns a subword tokenization end-to-end as part of the model. We introduce a soft gradient-based subword tokenization module (GBST) that automatically learns latent subword representations from characters. We additionally introduce Charformer, a deep Transformer model that integrates GBST and operates on the byte level.
  arXiv  Detail & Related papers  (2021-06-23T22:24:14Z)
• ByT5: Towards a token-free future with pre-trained byte-to-byte models [23.532359202069063]
  Most widely-used pre-trained language models operate on sequences of tokens corresponding to word or subword units. We show that a standard Transformer architecture can be used with minimal modifications to process byte sequences. We also demonstrate that byte-level models are significantly more robust to noise and perform better on tasks that are sensitive to spelling and pronunciation.
  arXiv  Detail & Related papers  (2021-05-28T07:03:22Z)
• Segatron: Segment-Aware Transformer for Language Modeling and Understanding [79.84562707201323]
  We propose a segment-aware Transformer (Segatron) to generate better contextual representations from sequential tokens. We first introduce the segment-aware mechanism to Transformer-XL, which is a popular Transformer-based language model. We find that our method can further improve the Transformer-XL base model and large model, achieving 17.1 perplexity on the WikiText-103 dataset.
  arXiv  Detail & Related papers  (2020-04-30T17:38:27Z)

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.