Dataset Viewer
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
behavior: string
elo_mean: double
transcript_id: string
pattern_id: string
n_tokens: int64
meta: struct<model_repo: string, n_lines: int64, gpu: string, generated_unix: int64>
child 0, model_repo: string
child 1, n_lines: int64
child 2, gpu: string
child 3, generated_unix: int64
entries: list<item: struct<text: string, ids: list<item: int64>, pieces: list<item: string>, results: list<it (... 90 chars omitted)
child 0, item: struct<text: string, ids: list<item: int64>, pieces: list<item: string>, results: list<item: struct< (... 78 chars omitted)
child 0, text: string
child 1, ids: list<item: int64>
child 0, item: int64
child 2, pieces: list<item: string>
child 0, item: string
child 3, results: list<item: struct<lines: list<item: string>, fve: list<item: double>, cos: list<item: double>>>
child 0, item: struct<lines: list<item: string>, fve: list<item: double>, cos: list<item: double>>
child 0, lines: list<item: string>
child 0, item: string
child 1, fve: list<item: double>
child 0, item: double
child 2, cos: list<item: double>
child 0, item: double
to
{'meta': {'model_repo': Value('string'), 'n_lines': Value('int64'), 'gpu': Value('string'), 'generated_unix': Value('int64')}, 'entries': List({'text': Value('string'), 'ids': List(Value('int64')), 'pieces': List(Value('string')), 'results': List({'lines': List(Value('string')), 'fve': List(Value('float64')), 'cos': List(Value('float64'))})})}
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
behavior: string
elo_mean: double
transcript_id: string
pattern_id: string
n_tokens: int64
meta: struct<model_repo: string, n_lines: int64, gpu: string, generated_unix: int64>
child 0, model_repo: string
child 1, n_lines: int64
child 2, gpu: string
child 3, generated_unix: int64
entries: list<item: struct<text: string, ids: list<item: int64>, pieces: list<item: string>, results: list<it (... 90 chars omitted)
child 0, item: struct<text: string, ids: list<item: int64>, pieces: list<item: string>, results: list<item: struct< (... 78 chars omitted)
child 0, text: string
child 1, ids: list<item: int64>
child 0, item: int64
child 2, pieces: list<item: string>
child 0, item: string
child 3, results: list<item: struct<lines: list<item: string>, fve: list<item: double>, cos: list<item: double>>>
child 0, item: struct<lines: list<item: string>, fve: list<item: double>, cos: list<item: double>>
child 0, lines: list<item: string>
child 0, item: string
child 1, fve: list<item: double>
child 0, item: double
child 2, cos: list<item: double>
child 0, item: double
to
{'meta': {'model_repo': Value('string'), 'n_lines': Value('int64'), 'gpu': Value('string'), 'generated_unix': Value('int64')}, 'entries': List({'text': Value('string'), 'ids': List(Value('int64')), 'pieces': List(Value('string')), 'results': List({'lines': List(Value('string')), 'fve': List(Value('float64')), 'cos': List(Value('float64'))})})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
WeirdChat × NLA explanations (Qwen3.6-27B)
Per-token natural-language-autoencoder (NLA) analysis of the highest-ranked WeirdChat transcripts. For each token of each transcript, an NLA actor (AV) verbalizes Qwen3.6-27B's layer-42 activation at that token into up to 10 salience-ordered explanation lines, and an NLA critic (AR) reconstructs the activation from the cumulative line prefixes — giving a fraction-of-variance- explained (FVE) curve over lines 1..k.
What's here
weirdchat_matryoshka_fve.parquet— flattened, one row per token position (24,290 rows) for the matryoshka (salience-ordering-trained) NLA.matryoshka/precache_weirdchat.json— the raw per-transcript nested form (34 entries;{text, ids, pieces, results:[{lines, fve, cos} | null]}).weirdchat_labels.json— per-transcript metadata (behavior, elo, ids).std/…— same for the standard (<explanation>-format) NLA (added when its run completes).
Selection
34 transcripts = top-3 by elo.mean for each of the 13 Qwen3.6-27B
behaviors in Transluce/WeirdChat
(subject_model = qwen/qwen3.6-27b; the pattern's highlight_transcript_id,
rendered with the Qwen chat template).
Parquet schema (one row per token position)
| column | meaning |
|---|---|
transcript_idx |
0..33, index of the transcript |
behavior |
WeirdChat behavior id (the category), e.g. laser-at-aircraft |
elo_mean |
pattern's mean Elo (interestingness rank within the behavior) |
transcript_id, pattern_id |
WeirdChat provenance |
position |
token index in the transcript |
token_piece |
the decoded token at this position |
n_lines |
# explanation lines the AV produced (0 = none/degenerate) |
lines |
list[str] — the AV's salience-ordered explanation lines |
fve |
list[float] — cumulative round-trip FVE using lines 1..k |
marginal_fve |
list[float] — first differences of fve (per-line contribution) |
cos |
list[float] — cosine(pred, gold) per prefix |
full_fve |
FVE using all lines (== fve[-1]) |
FVE = 1 − ‖n(v̂_k) − n(v)‖² / ‖n(v) − μ‖², both pred/gold L2-normalized to
mse_scale, μ = population mean of normalized held-out activations. Front-
loading shows up as a steep early marginal_fve.
Caveats
- ~11% of positions carry a stray CJK char in some line — the model's known minor-leak behavior on this adversarial text, faithfully captured (not a bug).
- Early positions (low left-context) produce more generic/degenerate output.
- Source: natural_language_autoencoders;
models
ceselder/nla-qwen36-27b-matryoshka(matryoshka),ceselder/qwen3.6-27b-nla-L42(std).
- Downloads last month
- -