Fix YAML: 1BT config now has train + test splits (exhaustive 4-digit)

README.md

@@ -1,18 +1,17 @@
 ---
 configs:
-- config_name: test
-  data_files:
-  - split: test
-    path: test/test-*.parquet
 - config_name: 1BT
   data_files:
   - split: train
     path: 1BT/train-*.parquet
+  - split: test
+    path: 1BT/test-*.parquet
 ---
 
-# 4-digit stratified addition dataset
+# 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.
 
-Stratified digit count 1..4 for each operand (mirrors the 9-digit `generate_addition.py`
-regime in base 10).  ~46M train examples (pool plateaued near the 50M-pair
-commutative-unique ceiling), ~404M tokens.  5K held-out test examples.
-Train/test split is commutative-safe.
+Train ~45M examples (~570M tokens under Llama-3 tokenizer); test ~5M examples.