Paper Detail

Compressing Sequences in the Latent Embedding Space: $K$-Token Merging for Large Language Models

Zihao Xu, John Harvill, Ziwei Fan, Yizhou Sun, Hao Ding, Hao Wang

arxiv Score 13.3

Published 2026-04-16 · First seen 2026-04-17

General AI

Abstract

Large Language Models (LLMs) incur significant computational and memory costs when processing long prompts, as full self-attention scales quadratically with input length. Token compression aims to address this challenge by reducing the number of tokens representing inputs. However, existing prompt-compression approaches primarily operate in token space and overlook inefficiencies in the latent embedding space. In this paper, we propose K-Token Merging, a latent-space compression framework that merges each contiguous block of K token embeddings into a single embedding via a lightweight encoder. The compressed sequence is processed by a LoRA-adapted LLM, while generation remains in the original vocabulary. Experiments on structural reasoning (Textualized Tree), sentiment classification (Amazon Reviews), and code editing (CommitPackFT) show that K-Token Merging lies on the Pareto frontier of performance vs. compression, achieving up to 75% input length reduction with minimal performance degradation.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
now
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@article{xu2026compressing,
  title = {Compressing Sequences in the Latent Embedding Space: \$K\$-Token Merging for Large Language Models},
  author = {Zihao Xu and John Harvill and Ziwei Fan and Yizhou Sun and Hao Ding and Hao Wang},
  year = {2026},
  abstract = {Large Language Models (LLMs) incur significant computational and memory costs when processing long prompts, as full self-attention scales quadratically with input length. Token compression aims to address this challenge by reducing the number of tokens representing inputs. However, existing prompt-compression approaches primarily operate in token space and overlook inefficiencies in the latent embedding space. In this paper, we propose K-Token Merging, a latent-space compression framework that m},
  url = {https://arxiv.org/abs/2604.15153},
  keywords = {cs.CL, cs.AI},
  eprint = {2604.15153},
  archiveprefix = {arXiv},
}

Metadata

{}