Multi-word sentences 1-100 tokens; remove dataset_info; validate digits fixed

README.md

@@ -1,28 +1,7 @@
----
-dataset_info:
-  features:
-  - name: prompt
-    dtype: json
-  - name: completion
-    dtype: json
-  - name: tools
-    dtype: string
-  - name: chat_template_kwargs
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 0
-    num_examples: 3000
-  - name: validation
-    num_bytes: 0
-    num_examples: 300
-task_categories:
-  - text-generation
-license: apache-2.0
----
-
 # Reverse Keep Numbers
 
-This dataset pairs inputs with outputs where non-digit characters are reversed while digits remain in-place.
+This dataset contains synthetic chat-style examples where the assistant reverses non-digit characters while keeping digits in-place and unchanged.
 
-Schema uses OpenAI-style chat messages in `prompt` and `completion`, compatible with Qwen chat templates.
+- Input format: OpenAI-style chat messages in `prompt` and `completion`.
+- Tokenization-agnostic: examples are whitespace-delimited tokens; each token is reversed independently (digits fixed).
+- Splits: `train` (3000 rows), `validation` (300 rows).