Upload generated eval data for xsum (with 'eval_size' of 320)

EVAL DATASET GENERATED FROM ORIGINAL DATASET'S TRAINING SPLIT (with 'eval_size' of 320)
Hard negative mining performed AFTER train/eval splitting original pool size: 60819; so, this eval split may contain negatives from training split)

Source dataset: bobox/xSum-processed
(dataset name: xsum).
Hard negatives mined with model: sentence-transformers/all-MiniLM-L6-v2.
Mining params: max_score=0.8, num_negatives=1, relative_margin=0.05, absolute_margin=0.03, range_max=100.

README.md

@@ -63,8 +63,11 @@ dataset_info:
   - name: train
     num_bytes: 129791569
     num_examples: 53830
-  download_size: 84029467
-  dataset_size: 129791569
+  - name: eval
+    num_bytes: 766870
+    num_examples: 320
+  download_size: 84535612
+  dataset_size: 130558439
 configs:
 - config_name: natural-questions-1HN
   data_files:
@@ -88,4 +91,6 @@ configs:
   data_files:
   - split: train
     path: xsum-1HN/train-*
+  - split: eval
+    path: xsum-1HN/eval-*
 ---

xsum-1HN/eval-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5983de9e586b8ba5d8edc8cb1666202f6a22244eff445b8324d31dc986f95d57
+size 506145