Datasets:

xushawn
/

t91-lgs

Error code:   StreamingRowsError
Exception:    TypeError
Message:      Couldn't cast array of type struct<commit: string, name: string> to null
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, 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 129, 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 544, 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 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, 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 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, 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 361, 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 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2315, in cast_table_to_schema
                  cast_array_to_feature(
                  ~~~~~~~~~~~~~~~~~~~~~^
                      table[name] if name in table_column_names else pa.array([None] * len(table), type=schema.field(name).type),
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                      feature,
                      ^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1861, in wrapper
                  return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
                                           ~~~~^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2155, in cast_array_to_feature
                  return array_cast(
                      array,
                  ...<2 lines>...
                      allow_decimal_to_str=allow_decimal_to_str,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 1863, in wrapper
                  return func(array, *args, **kwargs)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2016, in array_cast
                  raise TypeError(f"Couldn't cast array of type {_short_str(array.type)} to {_short_str(pa_type)}")
              TypeError: Couldn't cast array of type struct<commit: string, name: string> to null

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.

Leela Chess Zero T91 training games, lossless LGS store

All of Leela Chess Zero's test91 (T91) self-play training data, converted once into a compact, lossless columnar store (LGS). Every field of every lc0 V6 training record is kept: search values (root, best, played, original/raw network), moves-left estimates, visit counts, the full policy (as visit counts), the played and best moves, and the game result. Boards are not stored; replaying the played moves from each game's start position reproduces them, and the full V6 records, exactly.

The store is meant as the single source for building Stockfish-style binpacks and for other experiments, without downloading Leela's raw tarballs again.

Layout

shards/YYYY-MM/<tarball stem>.lgs: one shard per Leela hourly training tarball (training-run2-test91-YYYYMMDD-HHMM.tar on data.lczero.org). Each shard has a JSON sidecar with its converter summary and checksum.
docs/FORMAT.md: the format spec.
tools/: a Python reader (lgs.py) and validator.

Side files with extra per-position columns (Stockfish filter flags, relabels) may be added under side/ later; they never change the shards.

What is in a shard

One row per position (one lc0 training record), grouped by game:

Level	One row is	Holds
Game	one self-play game	game id, start position (Chess960 index per side; 518 = standard chess), result, how it ended
Position	one searched position	played and best move (lc0 policy index), visits, root/best/played/orig q and d, moves-left values, policy KLD, flags, raw result, Syzygy-rescored result, Syzygy WDL
Policy	one legal move of one position	visit count

About 50 bytes per position (zstd 19), versus about 203 in Leela's gzip files. Roughly 4% of T91 games are Chess960 (symmetric start positions); they are included.

Results: result_q is the real game result. tb_result is lc0's rescorer Syzygy-rescored result with deblunder off. Nothing in the store is deblundered.

Reading

import lgs  # tools/lgs.py; needs numpy and zstandard (python-chess for replay)
sh = lgs.Shard("shards/2026-01/training-run2-test91-20260116-1217.lgs")
best_q = sh.column("best_q")            # float32, one per position
for g in sh.games(limit=1):
    rp = g.replay()                      # boards via python-chess
    probs = g.probabilities(rp)          # exact 1858-float policies

A C++ header-only reader is in the converter's source tree (src/trainingdata/lgs_writer.h and the companion reader).

Provenance

Source: Leela Chess Zero test91 training data, ODbL.
Converter: lc0 rescorer store mode, fork xu-shawn/lc0, branch t91-store (commit recorded in every shard header), 6-piece Syzygy WDL and DTZ tablebases.
Each shard was checked on conversion: every record's board, castling, 50-move counter and legal-move set must match the replay. Games that can't be stored losslessly are listed in the shard header's skipped_games.

Downloads last month: 7,654

Total file size:

637 GB