ByteFlow: Language Modeling through Adaptive Byte Compression without a Tokenizer

TL;DR

ByteFlow Net introduces a tokenizer-free, adaptive byte compression architecture for language modeling, enabling models to learn their own segmentation of raw byte streams, leading to improved performance over traditional tokenization methods.

Contribution

It proposes a novel hierarchical model that removes the need for fixed tokenizers by learning segmentation based on compression, enhancing adaptability and effectiveness.

Findings

Outperforms BPE-based Transformers and previous byte-level models.

Demonstrates improved language modeling performance.

Shows that adaptive, compression-driven segmentation benefits downstream tasks.

Abstract

Modern language models still rely on fixed, pre-defined subword tokenizations. Once a tokenizer is trained, the LM can only operate at this fixed level of granularity, which often leads to brittle and counterintuitive behaviors even in otherwise strong reasoning models. We introduce \textbf{ByteFlow Net}, a new hierarchical architecture that removes tokenizers entirely and instead enables models to learn their own segmentation of raw byte streams into semantically meaningful units. ByteFlow Net performs compression-driven segmentation based on the coding rate of latent representations, yielding adaptive boundaries \emph{while preserving a static computation graph via Top-$K$ selection}. Unlike prior self-tokenizing methods that depend on brittle heuristics with human-designed inductive biases, ByteFlow Net adapts its internal representation granularity to the input itself. Experiments demonstrate that this compression-based chunking strategy yields substantial performance gains, with ByteFlow Net outperforming both BPE-based Transformers and previous byte-level architectures. These results suggest that end-to-end, tokenizer-free modeling is not only feasible but also more effective, opening a path toward more adaptive and information-grounded language models.