entity2260/cl-pncm-2.8b
Text Generation • 3B • Updated • 251 • 1
Pure-Transformer Next-Concept Models (C-LPNCM) with RVQ-FSQ, Brier epistemic routing, equilibrium loops, and pointer decoding.
Note 2.89B physical / 3.60B effective params — flagship pure-Transformer C-LPNCM (5.88 GB BF16 VRAM)
Note 124M compact C-LPNCM checkpoint for rapid experimentation on T4 / Colab
Note Foundational paper: Next-Concept Modeling: Fast, Parallel Coarse-to-Fine Text Generation (Kumbong et al., 2026)
Note Next-Concept Modeling training datasets