LoPT: Lossless Parallel Tokenization Acceleration for Long Context Inference of Large Language Model

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Shao, Wei, Zheng, Lingchao, Wang, Pengyu, Zheng, Peizhen, Li, Jun, Fan, Yuwei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911253380726784
author Shao, Wei
Zheng, Lingchao
Wang, Pengyu
Zheng, Peizhen
Li, Jun
Fan, Yuwei
author_facet Shao, Wei
Zheng, Lingchao
Wang, Pengyu
Zheng, Peizhen
Li, Jun
Fan, Yuwei
contents Long context inference scenarios have become increasingly important for large language models, yet they introduce significant computational latency. While prior research has optimized long-sequence inference through operators, model architectures, and system frameworks, tokenization remains an overlooked bottleneck. Existing parallel tokenization methods accelerate processing through text segmentation and multi-process tokenization, but they suffer from inconsistent results due to boundary artifacts that occur after merging. To address this, we propose LoPT, a novel Lossless Parallel Tokenization framework that ensures output identical to standard sequential tokenization. Our approach employs character-position-based matching and dynamic chunk length adjustment to align and merge tokenized segments accurately. Extensive experiments across diverse long-text datasets demonstrate that LoPT achieves significant speedup while guaranteeing lossless tokenization. We also provide theoretical proof of consistency and comprehensive analytical studies to validate the robustness of our method.
format Preprint
id arxiv_https___arxiv_org_abs_2511_04952
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoPT: Lossless Parallel Tokenization Acceleration for Long Context Inference of Large Language Model
Shao, Wei
Zheng, Lingchao
Wang, Pengyu
Zheng, Peizhen
Li, Jun
Fan, Yuwei
Computation and Language
Long context inference scenarios have become increasingly important for large language models, yet they introduce significant computational latency. While prior research has optimized long-sequence inference through operators, model architectures, and system frameworks, tokenization remains an overlooked bottleneck. Existing parallel tokenization methods accelerate processing through text segmentation and multi-process tokenization, but they suffer from inconsistent results due to boundary artifacts that occur after merging. To address this, we propose LoPT, a novel Lossless Parallel Tokenization framework that ensures output identical to standard sequential tokenization. Our approach employs character-position-based matching and dynamic chunk length adjustment to align and merge tokenized segments accurately. Extensive experiments across diverse long-text datasets demonstrate that LoPT achieves significant speedup while guaranteeing lossless tokenization. We also provide theoretical proof of consistency and comprehensive analytical studies to validate the robustness of our method.
title LoPT: Lossless Parallel Tokenization Acceleration for Long Context Inference of Large Language Model
topic Computation and Language
url https://arxiv.org/abs/2511.04952