Update dataset card: 3MT-3digit (a+b<=999, all single-token)

README.md

@@ -12,12 +12,12 @@ configs:
   data_files:
   - split: train
     path: 10BT/train-*.parquet
-- config_name: 6MT-3digit
+- config_name: 3MT-3digit
   data_files:
   - split: train
-    path: 6MT-3digit/train-*.parquet
+    path: 3MT-3digit/train-*.parquet
   - split: test
-    path: 6MT-3digit/test-*.parquet
+    path: 3MT-3digit/test-*.parquet
 ---
 
 # Addition Dataset
@@ -29,13 +29,13 @@ Addition problems in the format `{a} + {b} = {c}`.
 - **test**: 5K held-out evaluation examples (operands >= 10, i.e. min 2 digits)
 - **1BT**: ~85M training examples (~1 billion tokens under Llama-3 tokenizer)
 - **10BT**: ~850M training examples (~10 billion tokens)
-- **6MT-3digit**: Exhaustive 0-999 addition (all 1000x1000 = 1M ordered pairs, ~6.5M tokens). Symmetry-safe train/test split (10% test).
+- **3MT-3digit**: Exhaustive single-token addition: all (a, b) with a, b in [0, 999] and a+b <= 999. 500,500 ordered pairs, ~3M tokens. All of a, b, c are single tokens. Symmetry-safe train/test split (10% test).
 
 ## Deduplication
 
 - Commutative dedup: if `a + b = c` exists, `b + a = c` is excluded (1BT/10BT)
 - Test exclusion: both orderings of test-set pairs are excluded from train splits
-- 6MT-3digit: both orderings always in the same split (no commutative leakage)
+- 3MT-3digit: both orderings always in the same split (no commutative leakage)
 
 ## Usage
 
@@ -45,7 +45,7 @@ from datasets import load_dataset
 train = load_dataset("deqing/addition_dataset", "1BT", split="train")
 test = load_dataset("deqing/addition_dataset", "test", split="test")
 
-# 3-digit exhaustive (0-999)
-train_3d = load_dataset("deqing/addition_dataset", "6MT-3digit", split="train")
-test_3d = load_dataset("deqing/addition_dataset", "6MT-3digit", split="test")
+# Single-token exhaustive (0-999, a+b<=999)
+train_3d = load_dataset("deqing/addition_dataset", "3MT-3digit", split="train")
+test_3d = load_dataset("deqing/addition_dataset", "3MT-3digit", split="test")
 ```