fix training data

dataset.py

@@ -0,0 +1,38 @@
+mport os
+import json
+from datasets import DatasetInfo, GeneratorBasedBuilder, SplitGenerator, Split
+
+class AReaLBoba2RLCode(GeneratorBasedBuilder):
+    VERSION = "1.0.0"
+
+    def _info(self):
+        return DatasetInfo(
+            description="AReaL-boba-2-RL-Code dataset (only train split)",
+            features={
+                "input": {"dtype": "string", "id": None},
+                "output": {"dtype": "string", "id": None},
+            },
+            supervised_keys=None,
+        )
+
+    def _split_generators(self, dl_manager):
+        data_dir = dl_manager.download_and_extract(self.config.data_dir or ".")
+        train_path = os.path.join(data_dir, "train", "train.jsonl")
+
+        return [
+            SplitGenerator(
+                name=Split.TRAIN,
+                gen_kwargs={"filepath": train_path},
+            ),
+        ]
+
+    def _generate_examples(self, filepath):
+        with open(filepath, encoding="utf-8") as f:
+            for idx, line in enumerate(f):
+                if not line.strip():
+                    continue
+                data = json.loads(line)
+                yield idx, {
+                    "input": data.get("input", ""),
+                    "output": data.get("output", ""),
+                }

train/train.jsonl

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02c7abb1a272ad13e381390560363d668cca1b00c39ede7b08775e6fdf2dbe94
-size 6869752293
+oid sha256:07196b4fbc74e8961bc458261a5f1c361c9f7978482682218188937148342b3e
+size 6869950785