Deduplicate byte-identical documents (per chunk size)
Browse filesCollapses source_urls that are byte-identical full documents to a single canonical copy per documents-{N} config, and remaps query gold source_urls to the canonical (gold char spans unchanged). Makes nDCG@10 well defined (only one correct chunk per query).
queries/test-00000-of-00001.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0c70dd6cdf2a587638c2cdf781df44779bd971e723bc07b8fab6526868ef3890
|
| 3 |
+
size 548597
|