Language model predicts concepts to improve text generation efficiency
NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction
Computation and Language
Summary
Large language models usually learn to predict the next word in a sentence. This paper presents a model that also learns to predict larger chunks of meaning called concepts, which helps it understand and generate text better. The model builds a special set of concepts from its internal thoughts and uses these to guide the next word choices. This approach leads to better performance with less training data and can be adapted easily for different topics later on.
What this means in practice
- •For natural language processing teams: Reduce training data and computation needed to build large language models that generate more accurate text by predicting both tokens and concepts.
- •For ai system maintainers: Update a small module in a pretrained model to quickly adapt language generation to different domains without full retraining.
- •For enterprise ai developers: Build more efficient language-based tools that improve response length and quality by integrating latent concept predictions into generation pipelines.$Commercial implications: This paper enables novel commercial AI products that generate longer and more relevant responses using lightweight concept modules adaptable across sectors.
Authors
The Intern-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
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.