Hugging Face's logo

Datasets:
kaushikreddyxyz
/
corpus-scores-overflow

Size:
1B<n<10B
Tags:
interpretability
probing
gemma-2
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
concepts: list<item: string>
  child 0, item: string
families: list<item: string>
  child 0, item: string
layers: list<item: int64>
  child 0, item: int64
scores_shape: string
axes: struct<0: string, 1: string, 2: string>
  child 0, 0: string
  child 1, 1: string
  child 2, 2: string
main: struct<320: string, 321: string, 322: string, 323: string, 324: string, 325: string, 326: string, 32 (... 374 chars omitted)
  child 0, 320: string
  child 1, 321: string
  child 2, 322: string
  child 3, 323: string
  child 4, 324: string
  child 5, 325: string
  child 6, 326: string
  child 7, 327: string
  child 8, 328: string
  child 9, 329: string
  child 10, 330: string
  child 11, 331: string
  child 12, 332: string
  child 13, 333: string
  child 14, 334: string
  child 15, 335: string
  child 16, 336: string
  child 17, 337: string
  child 18, 338: string
  child 19, 339: string
  child 20, 340: string
  child 21, 341: string
  child 22, 342: string
  child 23, 343: string
  child 24, 344: string
  child 25, 345: string
  child 26, 346: string
  child 27, 347: string
  child 28, 348: string
  child 29, 349: string
  child 30, 350: string
  child 31, 351: string
  child 32, 352: string
  child 33, 353: string
  child 34, 354: string
  child 35, 355: string
overflow: struct<356: string, 357: string, 358: string, 359: string, 360: string, 361: string, 362: string>
  child 0, 356: string
  child 1, 357: string
  child 2, 358: string
  child 3, 359: string
  child 4, 360: string
  child 5, 361: string
  child 6, 362: string
cap_bytes: int64
to
{'cap_bytes': Value('int64'), 'main': {'320': Value('string'), '321': Value('string'), '322': Value('string'), '323': Value('string'), '324': Value('string'), '325': Value('string'), '326': Value('string'), '327': Value('string'), '328': Value('string'), '329': Value('string'), '330': Value('string'), '331': Value('string'), '332': Value('string'), '333': Value('string'), '334': Value('string'), '335': Value('string'), '336': Value('string'), '337': Value('string'), '338': Value('string'), '339': Value('string'), '340': Value('string'), '341': Value('string'), '342': Value('string'), '343': Value('string'), '344': Value('string'), '345': Value('string'), '346': Value('string'), '347': Value('string'), '348': Value('string'), '349': Value('string'), '350': Value('string'), '351': Value('string'), '352': Value('string'), '353': Value('string'), '354': Value('string'), '355': Value('string')}, 'overflow': {'356': Value('string'), '357': Value('string'), '358': Value('string'), '359': Value('string'), '360': Value('string'), '361': Value('string'), '362': Value('string')}}
because column names don't match
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 489, 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 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              concepts: list<item: string>
                child 0, item: string
              families: list<item: string>
                child 0, item: string
              layers: list<item: int64>
                child 0, item: int64
              scores_shape: string
              axes: struct<0: string, 1: string, 2: string>
                child 0, 0: string
                child 1, 1: string
                child 2, 2: string
              main: struct<320: string, 321: string, 322: string, 323: string, 324: string, 325: string, 326: string, 32 (... 374 chars omitted)
                child 0, 320: string
                child 1, 321: string
                child 2, 322: string
                child 3, 323: string
                child 4, 324: string
                child 5, 325: string
                child 6, 326: string
                child 7, 327: string
                child 8, 328: string
                child 9, 329: string
                child 10, 330: string
                child 11, 331: string
                child 12, 332: string
                child 13, 333: string
                child 14, 334: string
                child 15, 335: string
                child 16, 336: string
                child 17, 337: string
                child 18, 338: string
                child 19, 339: string
                child 20, 340: string
                child 21, 341: string
                child 22, 342: string
                child 23, 343: string
                child 24, 344: string
                child 25, 345: string
                child 26, 346: string
                child 27, 347: string
                child 28, 348: string
                child 29, 349: string
                child 30, 350: string
                child 31, 351: string
                child 32, 352: string
                child 33, 353: string
                child 34, 354: string
                child 35, 355: string
              overflow: struct<356: string, 357: string, 358: string, 359: string, 360: string, 361: string, 362: string>
                child 0, 356: string
                child 1, 357: string
                child 2, 358: string
                child 3, 359: string
                child 4, 360: string
                child 5, 361: string
                child 6, 362: string
              cap_bytes: int64
              to
              {'cap_bytes': Value('int64'), 'main': {'320': Value('string'), '321': Value('string'), '322': Value('string'), '323': Value('string'), '324': Value('string'), '325': Value('string'), '326': Value('string'), '327': Value('string'), '328': Value('string'), '329': Value('string'), '330': Value('string'), '331': Value('string'), '332': Value('string'), '333': Value('string'), '334': Value('string'), '335': Value('string'), '336': Value('string'), '337': Value('string'), '338': Value('string'), '339': Value('string'), '340': Value('string'), '341': Value('string'), '342': Value('string'), '343': Value('string'), '344': Value('string'), '345': Value('string'), '346': Value('string'), '347': Value('string'), '348': Value('string'), '349': Value('string'), '350': Value('string'), '351': Value('string'), '352': Value('string'), '353': Value('string'), '354': Value('string'), '355': Value('string')}, 'overflow': {'356': Value('string'), '357': Value('string'), '358': Value('string'), '359': Value('string'), '360': Value('string'), '361': Value('string'), '362': Value('string')}}
              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.

corpus-scores-overflow

Concept-probe detection scores for ClimbMix, layers 6/8/14 co-located per token. Overflow continuation — same format as the main repo. Shards here: 356-362. Main repo: kaushikreddyxyz/corpus-scores.

Files (per shard <sid>, ClimbMix shards 320-362)

  • scores_<sid>.npy — int8 [n_tokens, 3, 54]
    • axis 0: token (aligns 1:1 with tokens_<sid>.npy / docs_<sid>.jsonl)
    • axis 1: layer — 0 = layer 6, 1 = layer 8, 2 = layer 14
    • axis 2: concept — index into columns.json concepts[] (54 concepts, 7 families)
  • tokens_<sid>.npy — int32 [n_tokens] gemma tokenizer ids
  • docs_<sid>.jsonl — per-document metadata (token spans into the shard)

Provenance

Gold detection probes (best of ridge/DoM/LDA per concept-layer) from concept-probes-gemma2-2b gold_probes/layer06|08|14, applied to gemma-2-2b residual activations on ClimbMix shards 320-362 (~46.6M tokens/shard, ~2.0B total). Scores are standardized per probe over ClimbMix, then int8-quantized (clipped at ±4σ); dequantize with quant.json: score = int8 * scale[l][c] + zero[l][c]. DoM steering scores are NOT in this repo.

Usage — pick exactly ONE layer per model 

import numpy as np
det = np.load("scores_00320.npy")   # [n, 3, 54]
l8 = det[:, 1, :]                   # layer 8 scores, [n, 54]
Downloads last month
442
Total file size:
54.1 GB