Dataset Viewer
Auto-converted to Parquet Duplicate
book_id
stringlengths
7
22
split
stringclasses
1 value
source
stringclasses
4 values
text
stringlengths
188
1.75M
spans
listlengths
0
1.88k
I001D528B
train
reviewed_opf
"༄༅། །རང་བྱུང་པདྨའི་སྙིང་ཐིག་སྔོན་(...TRUNCATED)
[ { "id": "a1d8b208135a463fb3d380b74ee734fc", "start": 206248, "end": 207747 } ]
I0288940D
train
reviewed_opf
"༄༅། །མཉམ་མེད་རྒོད་ཚང་པའི་ཞལ་གདམས་(...TRUNCATED)
[ { "id": "8f6d413e7e534ec69894405b60c1fcb9", "start": 29981, "end": 30072 } ]
I07ECF0AE
train
reviewed_opf
"༄༅། །བོད་ཀྱི་བརྡ་དག་བྱིས་པའི་ལམ་མ(...TRUNCATED)
[ { "id": "0cf10473675f4aa087d5bd88bc35d42a", "start": 4849, "end": 4985 } ]
I08405360
train
reviewed_opf
"༄༅། །སྡེ་བདུན་མདོ་དང་བཅས་པའི་དགོང(...TRUNCATED)
[ { "id": "25cf49c0dd714862854a171fdf67ccb4", "start": 346793, "end": 346880 } ]
I0FCFA88F
train
reviewed_opf
"དྭགས་བརྒྱུད་སྒྲུབ་པའི་ཐེམ་སྐས་གདམ(...TRUNCATED)
[{"id":"c3fb7d2dc07b4ac59c69f05d836a19e9","start":92142,"end":92156},{"id":"b089221dac9f41af9c3dbb02(...TRUNCATED)
I100E7DAD
train
reviewed_opf
"དྭགས་བརྒྱུད་སྒྲུབ་པའི་ཐེམ་སྐས་གདམ(...TRUNCATED)
[{"id":"3107d562a46a42899f89af6ed54b60f6","start":188854,"end":189120},{"id":"5c7024c98d774c38b05789(...TRUNCATED)
I163178A0
train
reviewed_opf
"༄༅། །སངས་རྒྱས་ཀྱི་སྐུའི་རབ་ཏུ་དབྱ(...TRUNCATED)
[{"id":"1b201eef6028405cad6f67c21ee88dcb","start":6908,"end":6986},{"id":"5c9429a1b790402986c1a2b556(...TRUNCATED)
I19A08A51
train
reviewed_opf
"༄༅། །བདེ་གཤེགས་སྙིང་པོའི་བསྟོད་པ་(...TRUNCATED)
[{"id":"075c29fa6b9f453687f938a74dd4bfdb","start":14266,"end":14466},{"id":"fc8109fb97e742969fb3a5e6(...TRUNCATED)
I1CC06C39
train
reviewed_opf
"༄༅། །དྲི་ལན་སྐོར་རྨོངས་པའི་མུན་སེ(...TRUNCATED)
[{"id":"bc6029f4757e4963a4de2eb5e1c0a9de","start":248,"end":281},{"id":"7fabc23ff36f4a1c84eb6aaae568(...TRUNCATED)
I1DCBD5CC
train
reviewed_opf
"༄༅། །ཕྱག་རྒྱ་ཆེན་པོའི་སྨོན་ལམ་གྱི(...TRUNCATED)
[{"id":"dd5649552e10484fb5d04e0267728580","start":3858,"end":3868},{"id":"b3a49ac8590647589822cff924(...TRUNCATED)
End of preview. Expand in Data Studio

Bo-Yigchung-Dataset

762 Classical Tibetan books with character-offset spans marking Yigchung, the small-print interlinear note written under or beside the main line. It is the data of BDRC/Bo-Yigchung-Detection; see that card for how the model was trained on it.

Format

  • Config default: one row per book. Columns: book_id, split, source, text, spans (list of {id, start, end}).
  • Offsets are character offsets, inclusive on both ends: text[start:end + 1].
  • Config windowed_w8192_s4914: tokenized windows for jhu-clsp/mmBERT-base, one row per window. 8,192 tokens per window (2 special tokens and 8,190 content tokens), step 4,914 (overlap 3,276), last window of a book end-aligned. Columns: book_id, split, window_index, kind, token_start, token_end, char_start, char_end, own_start, own_end, input_ids, attention_mask, labels, source. Labels are 0 O, 1 B-YIGCHUNG, 2 I-YIGCHUNG; -100 marks the two special tokens of each window. Every content token of every window carries its label, including tokens that an earlier window also covers; own_start and own_end give the token range of the window.
  • The text has no ༼ ༽ bracket marks.

Splits

Book-level: a book is in one split only.

Split Books Spans Windows Text characters
train 601 22,388 8,574 60,977,652
validation 84 2,637 1,041 7,456,195
test 77 4,059 1,189 8,470,389
total 762 29,084 10,804 76,904,236

Books by source:

Source Train Validation Test Total
reviewed_opf 180 24 24 228
lotsawa 374 48 47 469
bdrc 35 4 5 44
nalanda 12 8 1 21

Test spans by source: nalanda 1,580, bdrc 1,570, reviewed_opf 848, lotsawa 61.

Data source and preparation

Source Books What it is
reviewed_opf 228 Reviewed OpenPecha books
lotsawa 469 Short texts from Lotsawa House
nalanda 21 Nalanda Vinaya volumes
bdrc 44 BDRC volumes

Steps: combine the four sources, remove duplicate books and passages from the Lotsawa, BDRC and reviewed_opf train sets only, split by book, tokenize with jhu-clsp/mmBERT-base and cut windows. The scripts that did this are not in the tibetan-layer-detection repository.

Annotation

Spans were annotated through OpenPecha. The reviewed_opf books come from the reviewed OpenPecha set; the other sources keep their own annotations. The published spans are the original annotations: the later annotator review of the five BDRC test books (see Known limits) is not applied to this dataset.

Known limits

  • The test split is dominated by two sources: nalanda (1 book, 1,580 spans) and bdrc (5 books, 1,570 spans) hold 78% of the test spans.
  • Lotsawa books are short and carry few spans (61 test spans in 47 books).
  • Some gold spans have boundary or head-mark inconsistencies; an annotator review of the 5 BDRC test books changed 1,570 spans to 1,297.
  • Data rights: texts come from BDRC, Lotsawa House and Nalanda. Check each source's terms before reuse; the licence field is a placeholder.

Citation

BDRC Yigchung layer detection dataset (Bo-Yigchung-Dataset), Buddhist Digital Resource Center, 2026.

Acknowledgements

Source texts were digitized and made available by the Buddhist Digital Resource Center (BDRC). We gratefully acknowledge BDRC. Annotations were prepared through OpenPecha with support from the Tsadra Foundation.

Downloads last month
181

Models trained or fine-tuned on BDRC/Bo-Yigchung-Dataset