feat(evaluate): allow loading local datasets for evaluation (#362)

packages/ragbits-evaluate/CHANGELOG.md

@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- Add local data loader (#334).
+
 ## 0.8.0 (2025-01-29)
 
 ### Changed

packages/ragbits-evaluate/src/ragbits/evaluate/dataloaders/local.py

@@ -0,0 +1,45 @@
+from typing import TypeAlias
+
+from datasets import Dataset, DatasetDict, IterableDataset, IterableDatasetDict, load_dataset
+
+from .base import DataLoader
+
+HFData: TypeAlias = DatasetDict | Dataset | IterableDatasetDict | IterableDataset
+
+
+class LocalDataLoader(DataLoader[DatasetDict]):
+    """
+    Local data loader.
+    """
+
+    AVAILABLE_BUILDERS = {
+        "json",
+        "csv",
+        "parquet",
+        "arrow",
+        "text",
+        "xml",
+        "webdataset",
+        "imagefolder",
+        "audiofolder",
+        "videofolder",
+    }
+
+    def __init__(self, path: str, split: str, builder: str) -> None:
+        self.path = path
+        self.split = split
+        self.builder = builder
+
+        if self.builder not in self.AVAILABLE_BUILDERS:
+            raise ValueError(
+                f"Unsupported builder '{self.builder}'. Available builders: {', '.join(self.AVAILABLE_BUILDERS)}"
+            )
+
+    async def load(self) -> DatasetDict:
+        """
+        Load the data from the local file.
+
+        Returns:
+            The loaded data.
+        """
+        return load_dataset(self.builder, data_files=self.path, split=self.split)