Add long split: 60k long-document summarization/keypoints/highlights

.gitattributes

@@ -59,3 +59,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
 data/train.json filter=lfs diff=lfs merge=lfs -text
+data/long.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -12,6 +12,8 @@ configs:
   data_files:
   - split: train
     path: data/train.parquet
+  - split: long
+    path: data/long.parquet
 dataset_info:
   features:
   - name: id
@@ -27,24 +29,35 @@ dataset_info:
   splits:
   - name: train
     num_examples: 269939
+  - name: long
+    num_examples: 60000
 ---
 
 # summarization-250k
 
-**article → summarization / keypoints / highlights**
+**document → summarization / keypoints / highlights**
 
-English news articles paired with a short condensation of the same text. Each row is one article and one response. `task` selects the response style: a summary, key points, or highlights.
+Each row is one source document and one condensation. `task` selects the response style: a summary, key points, or highlights.
+
+## Splits
+
+| split | n | source |
+|---|---:|---|
+| `train` | 269,939 | English news articles |
+| `long` | 60,000 | long-form English documents (scientific / technical papers, typically 8k–16k tokens) |
+
+Load a split with `load_dataset("domofon/summarization-250k", split="train")` or `split="long"`.
 
 ## Format
 
-JSON Lines (`data/train.json`) and Parquet (`data/train.parquet`) share the same columns. The Hub viewer uses Parquet.
+JSON Lines (`data/{split}.json`) and Parquet (`data/{split}.parquet`) share the same columns. The Hub viewer uses Parquet.
 
 | field | content |
 |---|---|
 | `system` | summarization-assistant instruction |
-| `user` | article text only |
+| `user` | document text only |
 | `assistant` | the condensation |
 | `task` | `summarization`, `keypoints`, or `highlights` |
 | `id` | example id |
 
-The user field is the article. It does not include extra prompts or chat-template tags.
+The user field is the document. It does not include extra prompts or chat-template tags.

data/long.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd1fb9a53ba5af0dd53a301a0944a1ada326a839a50d9184c1e6a42e81d57f90
+size 3028437915

data/long.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a341900f528f3073886d94d7b9e95eb021f033ea9b6f656903b79cf3f9b01097
+size 1001217200