Update README after 10BT regeneration

README.md

@@ -1,52 +1,40 @@
 ---
 configs:
-- config_name: 10BT
+- config_name: test
   data_files:
-  - split: train
-    path: 10BT/train-*.parquet
+  - split: test
+    path: test/test-*.parquet
 - config_name: 1BT
   data_files:
   - split: train
     path: 1BT/train-*.parquet
-- config_name: test
+- config_name: 10BT
   data_files:
-  - split: test
-    path: test/test-*
-dataset_info:
-  config_name: test
-  features:
-  - name: text
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 138036
-    num_examples: 5000
-  download_size: 104821
-  dataset_size: 138036
+  - split: train
+    path: 10BT/train-*.parquet
 ---
 
 # Addition Dataset
 
-Addition problems in the format `{a} + {b} = {c}
-` where operands are uniformly sampled from [0, 999999999].
+Addition problems in the format `{a} + {b} = {c}` where operands have stratified digit counts (1-9 digits, uniformly sampled).
 
 ## Subsets
 
-- **1BT**: ~69M training examples (~1 billion tokens under Llama-3 tokenizer)
-- **10BT**: ~690M training examples (~10 billion tokens)
-- **test**: 5K held-out evaluation examples
+- **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)
+
+## Deduplication
+
+- Commutative dedup: if `a + b = c` exists, `b + a = c` is excluded
+- Test exclusion: both orderings of test-set pairs are excluded from train splits
 
 ## Usage
 
 ```python
 from datasets import load_dataset
 
-# Load 1BT training data
 train = load_dataset("deqing/addition_dataset", "1BT", split="train")
-
-# Load 10BT training data
 train_10b = load_dataset("deqing/addition_dataset", "10BT", split="train")
-
-# Load test set
 test = load_dataset("deqing/addition_dataset", "test", split="test")
 ```