# Code for building an NCM Model (From Stanford HazyResearch Lab) — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/44e02b103f49fcc8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-11-19 11:21 ingested 2026-11-20 02:39

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1y06z8l/code_for_building_an_ncm_model_from_stanford/)

## retained text



```
HazyResearch hasn't shared the code they used to build their Next-Concept Model (NCM) from their paper, but I was able to build my own NCM model and replicate similar results to their original paper. 

Using this code, I was able to build a Japanese NCM model and recreate the fast text generation described in the paper. 


```



document fetched 2026-11-20 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1y06z8l/code_for_building_an_ncm_model_from_stanford
