Use datasets library for loading

Updates the Loading example to use load_dataset.

README.md

@@ -79,24 +79,15 @@ Because `label` and parts of `metadata` have mixed JSON types, preserve them as
 
 ## Loading
 
-Direct JSONL reading preserves every value exactly:
+Load the dataset with the Hugging Face `datasets` library:
 
 ```python
-import json
-from huggingface_hub import hf_hub_download
+from datasets import load_dataset
 
-path = hf_hub_download(
-    repo_id="wflying/rl-if-math-coding-logic-60k",
-    repo_type="dataset",
-    filename="rl-if-math-coding-logic-60k.jsonl",
+dataset = load_dataset(
+    "wflying/rl-if-math-coding-logic-60k",
+    split="train",
 )
-
-with open(path, encoding="utf-8") as file:
-    for line in file:
-        example = json.loads(line)
-        prompt = example["prompt"]
-        label = example["label"]
-        metadata = example["metadata"]
 ```
 
 Users who require `datasets.Dataset` should first normalize heterogeneous JSON fields or define an explicit schema appropriate for their training stack.