Deduplicated bitext_mining split(s) from kornwtp/alt-fil-khm-bitextmining (2026-08-28)

README.md

@@ -1,21 +1,23 @@
 ---
-dataset_info:
-  features:
-  - name: source
-    dtype: large_string
-  - name: target
-    dtype: large_string
-  - name: source_row_index
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 12169001
-    num_examples: 20061
-  download_size: 5238008
-  dataset_size: 12169001
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+tags:
+- sea-bed
+- deduplicated
 ---
+
+# alt-fil-khm-bitextmining
+
+Deduplicated copy of [`kornwtp/alt-fil-khm-bitextmining`](https://huggingface.co/datasets/kornwtp/alt-fil-khm-bitextmining),
+part of the SEA-BED data-quality work.
+
+- **Source dataset:** `kornwtp/alt-fil-khm-bitextmining`
+- **Deduplicated on:** 2026-08-28
+- **Task type:** bitext_mining
+- **Splits:** train
+
+## What changed
+
+Duplicate (source, target) pairs collapsed to one row, and per-side duplicate sources/targets collapsed so the split has a retrieval ceiling of 1.0 on both sides. Every removed row was an exact duplicate after normalization; no row was removed for being low quality.
+
+Row counts and the exact policy used are recorded in `bitext_mining_dedup_summary.csv` in the
+analysis repo. This dataset is *not* identical to the published SEA-BED data --
+scores computed on it are not directly comparable to published SEA-BED numbers.