Datasets:

matboz
/

qwen3.6-27b-agentic-misalignment-logs

Tags:

License:

Dataset card Files Files and versions

xet

Community

Dataset Viewer

Duplicate

Split (1)

train

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
label: string
results_dir: string
overall: struct<n: int64, harmful: int64, rate: double, n_missing_classification: int64>
  child 0, n: int64
  child 1, harmful: int64
  child 2, rate: double
  child 3, n_missing_classification: int64
by_scenario: struct<blackmail: struct<n: int64, harmful: int64, rate: double>, leaking: struct<n: int64, harmful: (... 22 chars omitted)
  child 0, blackmail: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 1, leaking: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
by_condition: struct<blackmail_explicit-america_replacement: struct<n: int64, harmful: int64, rate: double>, black (... 910 chars omitted)
  child 0, blackmail_explicit-america_replacement: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 1, blackmail_explicit-america_restriction: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 2, blackmail_explicit-none_replacement: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 3, blackmail_explicit-none_restriction: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
  
...
     child 2, rate: double
  child 8, leaking_explicit-none_replacement: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 9, leaking_explicit-none_restriction: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 10, leaking_none-none_replacement: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
  child 11, leaking_none-none_restriction: struct<n: int64, harmful: int64, rate: double>
      child 0, n: int64
      child 1, harmful: int64
      child 2, rate: double
status: string
start_time: string
total_samples: int64
failed_samples: int64
config: struct<prompts_dir: string, models: list<item: string>, samples_per_condition: int64, output_dir: st (... 219 chars omitted)
  child 0, prompts_dir: string
  child 1, models: list<item: string>
      child 0, item: string
  child 2, samples_per_condition: int64
  child 3, output_dir: string
  child 4, concurrency_limits: struct<vllm/qwen3: int64>
      child 0, vllm/qwen3: int64
  child 5, provider_concurrency_limits: struct<vllm: int64>
      child 0, vllm: int64
  child 6, classification_enabled: bool
  child 7, resume_from: null
  child 8, verbose: bool
  child 9, rate_limits: null
  child 10, debug: bool
  child 11, temperature: double
completed_samples: int64
experiment_id: string
to
{'experiment_id': Value('string'), 'start_time': Value('string'), 'config': {'prompts_dir': Value('string'), 'models': List(Value('string')), 'samples_per_condition': Value('int64'), 'output_dir': Value('string'), 'concurrency_limits': {'vllm/qwen3': Value('int64')}, 'provider_concurrency_limits': {'vllm': Value('int64')}, 'classification_enabled': Value('bool'), 'resume_from': Value('null'), 'verbose': Value('bool'), 'rate_limits': Value('null'), 'debug': Value('bool'), 'temperature': Value('float64')}, 'total_samples': Value('int64'), 'completed_samples': Value('int64'), 'failed_samples': Value('int64'), 'status': 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
              label: string
              results_dir: string
              overall: struct<n: int64, harmful: int64, rate: double, n_missing_classification: int64>
                child 0, n: int64
                child 1, harmful: int64
                child 2, rate: double
                child 3, n_missing_classification: int64
              by_scenario: struct<blackmail: struct<n: int64, harmful: int64, rate: double>, leaking: struct<n: int64, harmful: (... 22 chars omitted)
                child 0, blackmail: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 1, leaking: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
              by_condition: struct<blackmail_explicit-america_replacement: struct<n: int64, harmful: int64, rate: double>, black (... 910 chars omitted)
                child 0, blackmail_explicit-america_replacement: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 1, blackmail_explicit-america_restriction: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 2, blackmail_explicit-none_replacement: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 3, blackmail_explicit-none_restriction: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                
              ...
                   child 2, rate: double
                child 8, leaking_explicit-none_replacement: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 9, leaking_explicit-none_restriction: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 10, leaking_none-none_replacement: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
                child 11, leaking_none-none_restriction: struct<n: int64, harmful: int64, rate: double>
                    child 0, n: int64
                    child 1, harmful: int64
                    child 2, rate: double
              status: string
              start_time: string
              total_samples: int64
              failed_samples: int64
              config: struct<prompts_dir: string, models: list<item: string>, samples_per_condition: int64, output_dir: st (... 219 chars omitted)
                child 0, prompts_dir: string
                child 1, models: list<item: string>
                    child 0, item: string
                child 2, samples_per_condition: int64
                child 3, output_dir: string
                child 4, concurrency_limits: struct<vllm/qwen3: int64>
                    child 0, vllm/qwen3: int64
                child 5, provider_concurrency_limits: struct<vllm: int64>
                    child 0, vllm: int64
                child 6, classification_enabled: bool
                child 7, resume_from: null
                child 8, verbose: bool
                child 9, rate_limits: null
                child 10, debug: bool
                child 11, temperature: double
              completed_samples: int64
              experiment_id: string
              to
              {'experiment_id': Value('string'), 'start_time': Value('string'), 'config': {'prompts_dir': Value('string'), 'models': List(Value('string')), 'samples_per_condition': Value('int64'), 'output_dir': Value('string'), 'concurrency_limits': {'vllm/qwen3': Value('int64')}, 'provider_concurrency_limits': {'vllm': Value('int64')}, 'classification_enabled': Value('bool'), 'resume_from': Value('null'), 'verbose': Value('bool'), 'rate_limits': Value('null'), 'debug': Value('bool'), 'temperature': Value('float64')}, 'total_samples': Value('int64'), 'completed_samples': Value('int64'), 'failed_samples': Value('int64'), 'status': 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.

Qwen3.6-27B agentic-misalignment logs: base vs difficult-advice LoRA

Full rollout transcripts + judge classifications from the Anthropic agentic-misalignment honeypots (blackmail + leaking), run on:

qwen36_base/ — base Qwen/Qwen3.6-27B
qwen36_tulu/ — base + matboz/qwen3.6-27b-difficult-advice-tulu-lora (r=32)

12 conditions (2 scenarios x 3 goal-conflict settings x 2 urgency), 50 samples each = 600 rollouts per model. Judge: google/gemini-3-flash-preview. Thinking mode on.

Per-sample transcripts live at <model>/models/vllm_<name>/<condition>/sample_NNN/response.json, each with raw_response (full model output incl. chain-of-thought), metadata, and classification (harmful_behavior, classifier_verdict, classifier_reasoning). <model>/misalignment_summary.json holds the aggregated per-condition rates.

Headline

scenario	base	+ LoRA
blackmail	89.3%	34.3%
leaking	41.7%	16.3%
overall	65.5%	25.3%

Downloads last month: 31

Total file size:

73.2 MB