Hugging Face's logo

Datasets:
r4topunk
/
boar-packs

Languages:
English
Tags:
offline
rag
sqlite
openstreetmap
wikipedia
License:
Dataset card Files
xet
 Community
Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
id: string
sizeBytes: int64
sha256: string
pois: int64
vegan: int64
osmDate: timestamp[s]
url: string
lon: double
country: string
name: string
population: int64
bbox: string
tz: string
ascii: string
poolSize: int64
geonameid: int64
lat: double
to
{'id': Value('string'), 'geonameid': Value('int64'), 'name': Value('string'), 'ascii': Value('string'), 'lat': Value('float64'), 'lon': Value('float64'), 'country': Value('string'), 'population': Value('int64'), 'tz': Value('string'), 'bbox': Value('string'), 'poolSize': Value('int64')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              id: string
              sizeBytes: int64
              sha256: string
              pois: int64
              vegan: int64
              osmDate: timestamp[s]
              url: string
              lon: double
              country: string
              name: string
              population: int64
              bbox: string
              tz: string
              ascii: string
              poolSize: int64
              geonameid: int64
              lat: double
              to
              {'id': Value('string'), 'geonameid': Value('int64'), 'name': Value('string'), 'ascii': Value('string'), 'lat': Value('float64'), 'lon': Value('float64'), 'country': Value('string'), 'population': Value('int64'), 'tz': Value('string'), 'bbox': Value('string'), 'poolSize': Value('int64')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

BOAR offline knowledge packs

Data files for BOAR, an offline AI research app for iOS and Android. Each pack is a single SQLite file the app downloads (or imports from a file) and searches on the device, with no network. The app pins every file to the commit it was uploaded in (…/resolve/<commit-sha>/<path>) and checks its SHA-256 and size before use, so a file here never changes under an installed app: a rebuilt pack is a new commit.

Built by the scripts in the app's repository (scripts/build-*.mjs); see docs/KNOWLEDGE_PACKS.md and docs/POI_PACKS.md there.

Packs, sources and licenses

Each pack keeps the license of its sources. Share-alike packs stay share-alike; attribution travels with every document (per-article source URL and license in the pack's article_meta table, shown in the app next to each answer).

Path What Sources License
places/world-places.sqlite Gazetteer: cities and towns with 15,000+ people, to resolve "restaurants in Lisbon" to a place GeoNames cities15000 CC BY 4.0, © GeoNames
places/cities/<city>.sqlite Places to eat and drink around one of ten cities (below): name, address, opening hours, coordinates, diet tags (diet:vegan, diet:vegetarian), cuisine; plus Wikivoyage "Eat"/"Drink" listings OpenStreetMap via Geofabrik extracts; Wikivoyage ODbL 1.0, © OpenStreetMap contributors (database); CC BY-SA 4.0 (Wikivoyage text)
places/tiles/… The same places for the whole world as 1°×1° tiles (coming) as above as above
topics/boar-preparedness.sqlite Emergency and preparedness: first aid, survival, disasters, water, food preservation, self-sufficiency Wikipedia, Wikibooks (First Aid, Outdoor Survival), Wikivoyage, Appropedia; Ready.gov and the US National Park Service; US Army FM 21-76 Survival (1992) CC BY-SA 4.0 (wiki sources); public domain (US government works, 17 U.S.C. §105)
topics/boar-preparedness.manifest.json Every document in the preparedness pack with its source URL and license — —
topics/boar-crypto.sqlite Ethereum and cryptography: 3,141 documents 591 EIPs and 617 ERCs (ethereum/EIPs, ethereum/ERCs); 129 pages of the consensus specs, execution specs, execution APIs and Portal Network specs; the Ethereum Yellow Paper (27 sections); 228 ethereum.org pages; 180 Bitcoin BIPs whose header names a permissive license (BSD, MIT, CC0, public domain, CC BY; 33 without one left out); 1,369 Wikipedia articles from the cryptography, post-quantum, zero-knowledge, blockchain, Ethereum, cryptocurrency, smart contract and DeFi categories, relevance-filtered CC0 1.0 (EIPs, ERCs, specs); CC BY-SA 4.0 (Yellow Paper, Wikipedia); MIT (ethereum.org); each BIP's own license. Per document in the pack.
topics/boar-crypto.manifest.json Every document in the crypto pack with its source URL and license — —
wiki/en/boar-wiki-en-00.sqlite … -14.sqlite English Wikipedia in 15 packs (one per FineWiki shard; install any subset): 6,124,620 articles, 18.03 GB in all. The ~1M most-read articles in full, the rest as leads; Wikipedia redirects; lead embeddings of the most-read Wikipedia via HuggingFaceFW/finewiki (August 2025 dumps); redirects and title index from the enwiki dumps; one day of pageviews for ranking CC BY-SA 4.0
wiki/en/boar-wikivoyage-en.sqlite English Wikivoyage: 34,002 travel guides with their redirects and lead embeddings, 351 MB Wikivoyage dump CC BY-SA 4.0

Coming next: the world places tiles.

Which cities have a places pack

Ten metropolitan areas, each a box of about 20 km around the city (the 1°×1° world tiles, coming, cover everything else):

  • Chosen by hand as the first samples: São Paulo, Singapore, Taipei, Buenos Aires, Berlin.
  • Drawn at random, to test cities nobody picked: GeoNames cities15000 with a population of 1,000,000 or more (568 cities), minus the five above, sorted by GeoNames id, shuffled with Fisher-Yates and the mulberry32 generator, seed 20261003, first five taken (node scripts/draw-poi-cities.mjs --cities cities15000.txt --seed 20261003 --n 5 --exclude sao-paulo,singapore,taipei,buenos-aires,berlin). The draw gave Edmonton, Qujing, Queens, Biên Hòa and Ciudad Nezahualcóyotl.
City Places (OSM) Tagged vegan Wikivoyage listings OSM data as of
Edmonton 2,409 12 111 2026-09-26
Qujing 0 0 3 2026-09-26
Queens (with parts of Brooklyn and Manhattan) 19,963 665 48 2026-09-26
Biên Hòa (with eastern Ho Chi Minh City) 3,206 266 0 2026-09-26
Ciudad Nezahualcóyotl (with eastern Mexico City) 6,098 60 0 2026-09-26

OpenStreetMap has almost no restaurants mapped in Qujing, so its pack holds only the Wikivoyage listings, and the app says it has no offline data for places to eat there instead of making names up.

Attribution

  • OpenStreetMap: © OpenStreetMap contributors, available under the Open Database License (ODbL) 1.0, https://www.openstreetmap.org/copyright. The places packs are derived databases under the ODbL.
  • Wikipedia, Wikivoyage, Wikibooks, Appropedia: text under CC BY-SA 4.0; each document links to its source page, where the list of authors is.
  • GeoNames: CC BY 4.0, https://www.geonames.org.
  • US government works: public domain in the United States.

Nothing in these packs is hand-written about the topics the app is evaluated on.

File format

  • Topic and Wikipedia packs: "boar-knowledge-pack" format 2. Text in zstd-compressed blocks, an FTS5 keyword index over passages, redirects/aliases, and binary embeddings (bge-small-en-v1.5) of the leads of the most-read articles. The meta table records sources, build parameters and counts.
  • Places packs: SQLite with a places table, diet flags and the Wikivoyage listings; see docs/POI_PACKS.md in the app repository.

Sizes and SHA-256 of every file are in the app's catalog and in this repository's commit history.

Wave 1 (2026-10-12)

Seven packs built from the 2024-04-02 Stack Exchange data dump, the 2026-10-10 MedlinePlus Health Topics XML and the preparedness document set. Each folder holds the pack, its manifest, a README.md card, a LICENSE.md and a 20-item public selftest.jsonl. "User-selected only" means the pack passed the licence checks but displaced gold evidence from the default kit in the crowding probe, so the app should offer it as an opt-in, not install it by default. Stack Exchange packs stay CC BY-SA, with the per-post attribution inside each pack's article_meta.

Path Domain Size Licence Eligibility SHA-256 (first 8)
qa/qa-travel-se/ Travel (Stack Exchange Q&A) 141.5 MiB CC BY-SA 3.0/4.0 per post user-selected only a85521ce
qa/qa-ethereum-se/ Ethereum (Stack Exchange Q&A) 97.7 MiB CC BY-SA 3.0/4.0 per post user-selected only f16b258f
qa/qa-bitcoin-se/ Bitcoin (Stack Exchange Q&A) 74.3 MiB CC BY-SA 3.0/4.0 per post user-selected only 35b7a3ce
qa/qa-health-se/ Medical sciences (Stack Exchange Q&A) 22.7 MiB CC BY-SA 3.0/4.0 per post default-eligible 687e4994
qa/qa-outdoors-se/ Outdoors (Stack Exchange Q&A) 30.1 MiB CC BY-SA 3.0/4.0 per post default-eligible ff733e4f
topics/boar-firstaid-en/ First aid and emergency preparedness 20.0 MiB CC BY-SA 4.0 + public domain default-eligible 8cfab0b4
topics/boar-medlineplus-en/ MedlinePlus health topics 4.1 MiB Public domain (NLM) default-eligible c9ac4dcf

Wave 2 (2026-10-13)

Two general-reference packs: the CIA World Factbook CC0 archive (one entry per country or region, Key facts block + narrative) and the whole Simple English Wikipedia (dump 2026-10-01). Both are user-selected packs, not defaults: Simple English Wikipedia crowds the context on general questions, and the Factbook's country pages compete with the travel guides. Each folder holds the pack, its manifest, README, LICENSE and a 20-question public self-test.

Path Domain Size Licence Eligibility SHA-256 (first 8)
topics/boar-factbook/ CIA World Factbook (CC0 archive) 11.1 MiB CC0 (public domain) user-selected only c5854353
topics/boar-simplewiki-en/ Simple English Wikipedia (2026-10-01) 458.3 MiB CC BY-SA 4.0 + GFDL user-selected only c751768b

Wave 3 (2026-10-13)

Two more question-shaped packs from the 2024-04-02 Stack Exchange data dump: Seasoned Advice (cooking) and Personal Finance & Money. Both sites are old enough to carry CC BY-SA 2.5 posts next to 3.0 and 4.0; the per-post licence and attribution are in each pack's article_meta. The cooking pack passed the crowding probe (default-eligible); the money pack displaced gold evidence from the default kit on three benchmark items, so it is user-selected only.

Path Domain Size Licence Eligibility SHA-256 (first 8)
qa/qa-cooking-se/ Cooking (Seasoned Advice, Stack Exchange Q&A) 84.9 MiB CC BY-SA 2.5/3.0/4.0 per post default-eligible 81990fa9
qa/qa-money-se/ Personal finance (Stack Exchange Q&A) 130.9 MiB CC BY-SA 2.5/3.0/4.0 per post user-selected only 37411234
Downloads last month
499
Total file size:
20.7 GB