Papers/2609.10715
🧪 Test?View on arXiv

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

Not specified in the provided content

latent-space modelsnext-token predictionconcept prediction
2609.10715
Builder Relevance
80%
4h ago

Abstract

NCP-ArchPreview is a latent-space language model that enhances autoregressive pretraining by incorporating Next Concept Prediction alongside standard next-token prediction.

Reality Card

Core Claim

NCP-ArchPreview achieves a significant performance improvement over OLMo-3-7B by utilizing only 51.3% of the total training tokens and 85% of the standard computation.

Method / Result

Outperforms OLMo-3-7B by 2.45 points on the downstream macro-average, with a notable 5.99-point gain on GSM8K.

Limitations

The paper does not specify authors or provide detailed reproducibility instructions.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers