---
configs:
- config_name: 1BT
  data_files:
  - split: train
    path: 1BT/train-*.parquet
  - split: test
    path: 1BT/test-*.parquet
---

# 4-digit addition dataset (a + b = c, all <= 4 digits)

Exhaustive enumeration of every (a, b) with a, b in [0, 9999] and
a + b <= 9999 (so c is also <= 4 digits). 50,005,000 ordered pairs,
commutative-safe 90/10 train/test split.

Train ~45M examples (~570M tokens under Llama-3 tokenizer); test ~5M examples.