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
incidents: list<item: struct<entity: string, truth: struct<kind: string, playbook: string, family: string, gold (... 764 chars omitted)
child 0, item: struct<entity: string, truth: struct<kind: string, playbook: string, family: string, gold: string, w (... 752 chars omitted)
child 0, entity: string
child 1, truth: struct<kind: string, playbook: string, family: string, gold: string, what: string, cohort: int64>
child 0, kind: string
child 1, playbook: string
child 2, family: string
child 3, gold: string
child 4, what: string
child 5, cohort: int64
child 2, evidence: struct<source: string, events_seen: int64, events_alarmed: int64, entity_risk: double, families_name (... 612 chars omitted)
child 0, source: string
child 1, events_seen: int64
child 2, events_alarmed: int64
child 3, entity_risk: double
child 4, families_named: struct<attack_generic: int64, bruteforce: int64, dos: int64, portscan: int64, ddos: int64, bot: int6 (... 14 chars omitted)
child 0, attack_generic: int64
child 1, bruteforce: int64
child 2, dos: int64
child 3, portscan: int64
child 4, ddos: int64
child 5, bot: int64
child 6, rce: int64
child 5, ports_or_paths: list<item: int64>
child 0, item: int64
child 6, alarms: list<item: struct<kind: string, fam
...
int64, distinct_sources: (... 74 chars omitted)
child 0, destination: string
child 1, port: int64
child 2, flows_in_window: int64
child 3, window_s: int64
child 4, distinct_sources: int64
child 5, share_unanswered: double
child 6, destination_normal_flows_per_min: double
child 10, site_pressure: null
system_prompt: string
entity: string
reasoning: string
beam_action: string
miss: int64
gold: string
effort: string
exact: int64
usage: struct<prompt_tokens: int64, completion_tokens: int64, total_tokens: int64, prompt_tokens_details: s (... 200 chars omitted)
child 0, prompt_tokens: int64
child 1, completion_tokens: int64
child 2, total_tokens: int64
child 3, prompt_tokens_details: struct<audio_tokens: int64, cached_tokens: int64>
child 0, audio_tokens: int64
child 1, cached_tokens: int64
child 4, completion_tokens_details: struct<accepted_prediction_tokens: int64, audio_tokens: int64, reasoning_tokens: int64, rejected_pre (... 22 chars omitted)
child 0, accepted_prediction_tokens: int64
child 1, audio_tokens: int64
child 2, reasoning_tokens: int64
child 3, rejected_prediction_tokens: int64
seconds: double
harm: int64
answer: struct<explanation: string, confidence: double, action: string, reason: string>
child 0, explanation: string
child 1, confidence: double
child 2, action: string
child 3, reason: string
kind: string
to
{'effort': Value('string'), 'entity': Value('string'), 'kind': Value('string'), 'gold': Value('string'), 'beam_action': Value('string'), 'exact': Value('int64'), 'harm': Value('int64'), 'miss': Value('int64'), 'answer': {'explanation': Value('string'), 'confidence': Value('float64'), 'action': Value('string'), 'reason': Value('string')}, 'reasoning': Value('string'), 'usage': {'prompt_tokens': Value('int64'), 'completion_tokens': Value('int64'), 'total_tokens': Value('int64'), 'prompt_tokens_details': {'audio_tokens': Value('int64'), 'cached_tokens': Value('int64')}, 'completion_tokens_details': {'accepted_prediction_tokens': Value('int64'), 'audio_tokens': Value('int64'), 'reasoning_tokens': Value('int64'), 'rejected_prediction_tokens': Value('int64')}}, 'seconds': 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 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 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
incidents: list<item: struct<entity: string, truth: struct<kind: string, playbook: string, family: string, gold (... 764 chars omitted)
child 0, item: struct<entity: string, truth: struct<kind: string, playbook: string, family: string, gold: string, w (... 752 chars omitted)
child 0, entity: string
child 1, truth: struct<kind: string, playbook: string, family: string, gold: string, what: string, cohort: int64>
child 0, kind: string
child 1, playbook: string
child 2, family: string
child 3, gold: string
child 4, what: string
child 5, cohort: int64
child 2, evidence: struct<source: string, events_seen: int64, events_alarmed: int64, entity_risk: double, families_name (... 612 chars omitted)
child 0, source: string
child 1, events_seen: int64
child 2, events_alarmed: int64
child 3, entity_risk: double
child 4, families_named: struct<attack_generic: int64, bruteforce: int64, dos: int64, portscan: int64, ddos: int64, bot: int6 (... 14 chars omitted)
child 0, attack_generic: int64
child 1, bruteforce: int64
child 2, dos: int64
child 3, portscan: int64
child 4, ddos: int64
child 5, bot: int64
child 6, rce: int64
child 5, ports_or_paths: list<item: int64>
child 0, item: int64
child 6, alarms: list<item: struct<kind: string, fam
...
int64, distinct_sources: (... 74 chars omitted)
child 0, destination: string
child 1, port: int64
child 2, flows_in_window: int64
child 3, window_s: int64
child 4, distinct_sources: int64
child 5, share_unanswered: double
child 6, destination_normal_flows_per_min: double
child 10, site_pressure: null
system_prompt: string
entity: string
reasoning: string
beam_action: string
miss: int64
gold: string
effort: string
exact: int64
usage: struct<prompt_tokens: int64, completion_tokens: int64, total_tokens: int64, prompt_tokens_details: s (... 200 chars omitted)
child 0, prompt_tokens: int64
child 1, completion_tokens: int64
child 2, total_tokens: int64
child 3, prompt_tokens_details: struct<audio_tokens: int64, cached_tokens: int64>
child 0, audio_tokens: int64
child 1, cached_tokens: int64
child 4, completion_tokens_details: struct<accepted_prediction_tokens: int64, audio_tokens: int64, reasoning_tokens: int64, rejected_pre (... 22 chars omitted)
child 0, accepted_prediction_tokens: int64
child 1, audio_tokens: int64
child 2, reasoning_tokens: int64
child 3, rejected_prediction_tokens: int64
seconds: double
harm: int64
answer: struct<explanation: string, confidence: double, action: string, reason: string>
child 0, explanation: string
child 1, confidence: double
child 2, action: string
child 3, reason: string
kind: string
to
{'effort': Value('string'), 'entity': Value('string'), 'kind': Value('string'), 'gold': Value('string'), 'beam_action': Value('string'), 'exact': Value('int64'), 'harm': Value('int64'), 'miss': Value('int64'), 'answer': {'explanation': Value('string'), 'confidence': Value('float64'), 'action': Value('string'), 'reason': Value('string')}, 'reasoning': Value('string'), 'usage': {'prompt_tokens': Value('int64'), 'completion_tokens': Value('int64'), 'total_tokens': Value('int64'), 'prompt_tokens_details': {'audio_tokens': Value('int64'), 'cached_tokens': Value('int64')}, 'completion_tokens_details': {'accepted_prediction_tokens': Value('int64'), 'audio_tokens': Value('int64'), 'reasoning_tokens': Value('int64'), 'rejected_prediction_tokens': Value('int64')}}, 'seconds': 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.
Beam as the rule interpreter and incident analyst of a self-hosted security layer
Evaluation traces behind the Build with Beam entry DibaGuard x Beam (Space: https://huggingface.co/spaces/DibaAi/beam-dibaguard - model: https://huggingface.co/DibaAi/DibaGuard-Open). Model: reflection/Beam-1.0-501B-A23B through Hugging Face Inference Providers (OpenAI-compatible chat, reasoning_effort). Every row carries the prompt, Beam's answer, its reasoning_content, token usage, latency and the scores.
1. Rules in any language -> a precise rule (rules_bench.jsonl)
An operator writes a protection rule in plain language; the layer needs a deterministic predicate over what it measures (ports, paths, user-agents, rates, failure shares, CIDRs). 12 intents x 8 languages (en, es, de, fr, tr, ja, ar, hi); gold rule and 8 test contexts per intent (rules_intents.json). Behaviour agreement = the gold rule and Beam's rule take the same decision on the same contexts (match and action); exact = all 8 agree. The built-in parser (English plus one other language, no model) is the baseline.
| interpreter | n | behaviour agreement | exact rules | action right | scope right | gold conditions recovered | unusable answers | s / rule |
|---|---|---|---|---|---|---|---|---|
| built-in parser | 96 | 0.538 | 0.094 | 0.229 | 0.469 | 0.352 | 22 | 0.0 |
| Beam, effort low | 96 | 0.990 | 0.917 | 1.000 | 0.844 | 0.972 | 0 | 1.8 |
| Beam, effort medium | 96 | 0.984 | 0.906 | 0.990 | 0.833 | 0.972 | 0 | 2.6 |
| Beam, effort high | 96 | 0.977 | 0.906 | 0.969 | 0.823 | 0.948 | 3 | 2.6 |
Per language (behaviour agreement):
| interpreter | en | es | de | fr | tr | ja | ko | hi |
|---|---|---|---|---|---|---|---|---|
| parser | 0.74 | 0.51 | 0.55 | 0.49 | 0.51 | 0.49 | 0.54 | 0.47 |
| low | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 |
| medium | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.99 | 0.95 | 0.99 |
| high | 0.99 | 0.96 | 0.96 | 0.99 | 0.99 | 0.99 | 0.99 | 0.95 |
2. The analyst: evidence -> explanation + one action (analyst_bench.jsonl)
Incidents produced by the open build (licence-clean network models, no HTTP request model) over playbook attacks and over benign decoys that trip behavioural tiers (a partner API poller, the nightly backup host, a new sync job, a person on stale links, a declared partner integration). Beam gets the engine's own evidence for one entity and must propose exactly one action. Gold: the policy's action for an attack family; monitor for a decoy. harm = a decoy Beam would block or rate-limit; miss = an attack Beam would only monitor.
| effort | incidents | exact action | harm (of decoys) | miss (of attacks) | s / incident |
|---|---|---|---|---|---|
| low | 13 | 7/13 | 0/5 | 0/8 | 3.0 |
| medium | 13 | 9/13 | 0/5 | 0/8 | 3.4 |
| high | 13 | 7/13 | 0/5 | 1/8 | 5.2 |
3. Cost
Rules: 36 calls, 15423 prompt + 15434 completion tokens (13327 reasoning). Analyst: 39 calls, 30064 prompt + 41495 completion tokens (36864 reasoning).
4. How to reproduce
python scripts/beam_rules_bench.py --efforts low,medium,high
python scripts/beam_analyst_bench.py --efforts low,medium,high
python scripts/make_beam_dataset.py --push
(scripts in the model repo's runtime/ tree; HF_TOKEN or REFLECTION_API_KEY in the environment)
5. Licence
CC BY 4.0. The prompts, gold rules, contexts and incident generators are original; Beam's outputs are included as produced.
- Downloads last month
- 96