K2 Horizon Open-Sources 13 Trillion Token Pretraining Data and Pipelines

IFM, the K2 team, has open-sourced K2 Horizon's full pretraining and post-training data pipelines, billed as the most detailed open LLM data release to date, covering 13 trillion tokens across 5 datasets and 24 subsets, plus synthetic data pipelines exceeding 10 trillion tokens.

2026-10-24 ~ 2026-10-24 · 2 related posts

• K2 Horizon opens 13T pretraining tokens and post-training data pipelines, most detailed yet — ChengleiSi · 2026-10-24

1 near-duplicate retellings: SinclairWang1