refactor Trainer train/val loop

example/dataset.py

@@ -0,0 +1,20 @@
+from typing import NamedTuple
+
+from trainlib.domain import SequenceDomain
+from trainlib.datasets.memory import TupleDataset
+
+
+class Record(NamedTuple):
+    a: int
+    b: str
+
+tl_domain = SequenceDomain[Record]([
+    Record(1, "1"),
+    Record(2, "2"),
+])
+
+class R0(TupleDataset[Record]):
+    item_tuple = Record
+
+    def _process_item_data(self, item_data, item_index):
+        return (item_data[0],)