Skip to content

Dataset · IFM

Large text corpus for pretraining with a behavioral focus

IFM/Pretrain-Behaviors

Array of texts from the K2 Horizon collection with 1-10B records for training models in reasoning, planning, and format handling.

What's inside

The repository collects text data from the K2 Horizon series covering reasoning, planning, data science, gaming, general content, and format rewriting. The volume ranges from 1 billion to 10 billion examples. The data is divided into thematic subsets and stored in parquet format with streaming support.

What you can do with it

  • Pretraining and fine-tuning LLMs in logical reasoning and planning skills
  • Improving model capability to follow complex instructions and change text format
Tasks
Text generation
Size
1B–10B rows
Downloads this month
69.6K
Likes
42

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("IFM/Pretrain-Behaviors")