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:    TypeError
Message:      Couldn't cast array of type
struct<base_model.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-base-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-s1pp-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-s2pp-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>>
to
{'base_model.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}, 'deepseek-base-epoch_1.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'no
...
'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}, 'deepseek-s2pp-epoch_1.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}}
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 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, 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 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2312, 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 2158, in cast_array_to_feature
                  raise TypeError(f"Couldn't cast array of type\n{_short_str(array.type)}\nto\n{_short_str(feature)}")
              TypeError: Couldn't cast array of type
              struct<base_model.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-base-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-s1pp-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>, gemma4-s2pp-epoch_1.judged.json: struct<meta: struct<base: string, adapter: string, tokenizer: string, system_in_user: bool, thinking: string, n: int64, tokens: int64, capped: int64, empty_reasoning: int64, classes: struct<CONCEAL: int64, DISCLOSE: int64, EVASIVE: int64, ERROR: int64>, t_gen_s: int64, judge: string>, summary: struct<n_positive: int64, conceal: int64, disclose: int64, evasive: int64, detected: int64, not_detected: int64, deception_rate: double, detection_rate: double, undetected_rate: double, monitored: int64, judge: string, t_judge_s: int64, adapter: string>>>
              to
              {'base_model.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}, 'deepseek-base-epoch_1.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'no
              ...
              '), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}, 'deepseek-s2pp-epoch_1.judged.json': {'meta': {'base': Value('string'), 'adapter': Value('string'), 'tokenizer': Value('string'), 'system_in_user': Value('bool'), 'thinking': Value('string'), 'n': Value('int64'), 'tokens': Value('int64'), 'capped': Value('int64'), 'empty_reasoning': Value('int64'), 'classes': {'CONCEAL': Value('int64'), 'DISCLOSE': Value('int64'), 'EVASIVE': Value('int64'), 'ERROR': Value('int64')}, 't_gen_s': Value('int64'), 'judge': Value('string')}, 'summary': {'n_positive': Value('int64'), 'conceal': Value('int64'), 'disclose': Value('int64'), 'evasive': Value('int64'), 'detected': Value('int64'), 'not_detected': Value('int64'), 'deception_rate': Value('float64'), 'detection_rate': Value('float64'), 'undetected_rate': Value('float64'), 'monitored': Value('int64'), 'judge': Value('string'), 't_judge_s': Value('int64'), 'adapter': Value('string')}}}

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.

SDF model organisms: evaluation results

Result files behind every figure of the study, one folder per experiment. Each file is the raw output of the corresponding script in the code repository CoT-Verse (per-item records with the model's answer, scores and, where a chain of thought was generated, the trace), plus figure_materials/, the per-figure numbers the analysis scripts derive from them.

folder experiment
cot3d_A1_belief_rate single-fact multiple-choice belief rate, every organism of the dose array
cot3d_A3_leak free-generation leak of the implanted facts by domain
cot3d_B4_multihop multi-hop items with and without the model's own reasoning
rq04_dislodging inference-time pressure templates on both twins, durability pairs
rq05_cot_use reasoning-slot manipulations on the clean, false-fact and true-fact twins
rq06_read_the_cot belief-aware and belief-blind readers over the multi-hop traces (rq6_traces = the trace bundle)
rq07_grpo GRPO runs with a distorted-CoT reward: judge scores, checkpoint curves, reward logs
rq08_monitoring_notice the same items under a plain prompt and a monitoring notice
rq09_concealment defect-concealment organisms: full evaluations, epoch sweeps, controllability suite
rq10_monitors the published monitors on the RQ8 traces: readers, Chen pairs, Thought Branches resampling
rq11_edit_boundary own-trace sentence edits by type, dose and position
rq12_named_forms Arcuschin / Chen definitions applied to the organism's traces; hint arm
rq14_localisation the SDF weight delta switched off and on per block of layers (ablation, isolation, bisection)
rq15_belief_switch private / monitored notices on the concealment organisms, opacity judgements

Model identifiers inside the files point at the organisms under the PS4CoT profile.

Downloads last month
23