Initial siRBench dataset splits

.gitignore

@@ -0,0 +1,2 @@
+.DS_Store
+Thumbs.db

README.md

@@ -0,0 +1,32 @@
+---
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: siRBench_train.csv
+  - split: test
+    path: siRBench_test.csv
+  - split: leftout
+    path: siRBench_leftout.csv
+---
+
+# siRBench data
+
+This repository contains the three public siRBench tabular splits used by the
+model and benchmark workflows.
+
+Splits:
+- `train`: 2,776 rows
+- `test`: 275 rows
+- `leftout`: 896 rows
+
+All three files share the same 106-column schema, including:
+- `siRNA`
+- `mRNA`
+- `extended_mRNA`
+- `efficiency`
+- `source`
+- `cell_line`
+- engineered thermodynamic features
+
+`leftout` is a second held-out test set rather than a validation split.