Skip to content
AI Atlas
PaperActive

NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

arxiv.org/abs/2609.10715

Explore graph
quality84

Updated 4 h ago · first seen 11 Sept 2026

paper_01M294G4AZZMH57A5M561GXW5D

Published
11 Sept 2026
T1 · 5 h ago
arXiv
2609.10715
T1 · 5 h ago
Category
cs.CL
T1 · 5 h ago

Abstract

We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and more challenging concept-level objective while preserving standard token-level autoregressive generation. NCP-ArchPreview builds a latent space by constructing a product-quantized concept vocabulary directly from its hidden states, and subsequently learns to predict future concepts via a dedicated Concept Module. These predicted concepts are then fed back to the token level to guide subsequent generation, with NTP and NCP trained jointly end-to-end. We scale this architecture to 8.9B parameters and train it on 5.73T tokens from the Dolma-3 dataset, marking the largest demonstration of a latent-space language model to date. Remarkably, by consuming only 51.3% of the total training tokens, NCP-ArchPreview achieves the final pretraining loss of OLMo-3-7B. Following full pretraining, it outperforms OLMo-3-7B by 2.45 points on the downstream macro-average, including a notable 5.99-point gain on GSM8K. Controlled experiments isolate a clear progression of performance gains stemming from both the latent architecture and the NCP objective. Furthermore, utilizing only 85% of the standard computation, NCP-ArchPreview approaches the training loss of a strictly parameter-aligned 8.9B baseline. The learned latent space remains highly valuable after the pretraining stage: updating just the 17M-parameter VQ module yields a novel, lightweight interface for domain adaptation, while a simple injection of concept representations into a DFlash2 drafter improves the mean accepted length by 4.17% with negligible overhead.

Authors 28

NCP Team, Jiaqi Cao, Chiyu Chen, Shuang Cheng, Xu Cheng, Beiya Dai, Yufan Feng, Kewen Ge, Ruijun Ge, Jiayi Huang, Yang Jiao, Dahua Lin, Zhouhan Lin, Yifan Liu, Yuliang Liu, Biqing Qi, Mowen Ruan, Junzhe Shen, Yunchong Song, Hao Sun, Zhongbo Tian, Yixuan Wang, Rubin Wei, Jiaxin Xiong, Kangyu Yang, Qian Yao, Qi Zhang, Bowen Zhou

Specification

Official page
arxiv.org/abs/2609.10715

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Arxiv announce type
new

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

arXiv id
2609.10715

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Categories
cs.CL

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Hf paper url
https://huggingface.co/papers/2609.10715

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium

Hf comments
1

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium

Upvotes
165

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium

PDF
arxiv.org/pdf/2609.10715

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Primary category
cs.CL

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Published
11 Sept 2026

Source:arXiv (Atom API + RSS)T1observed 5 h agohigh

Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →

Provenance

Attributed facts

12

Source tiers

T1T29 / 3

Freshest observation

4 h ago

Conflicts

2 flagged