Datasets:

OsaurusAI
/

raptor-0.5-harness-corpus

Languages:

English

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
assistant_tokens: int64
benchmark_content_used: bool
capability: string
closure_kind: string
contamination_state: string
contrast_group_id: string
difficulty_band: string
effective_tokens: int64
effective_tokens_denominator: int64
effective_tokens_numerator: int64
execution_language: string
id: string
intent_class: string
interface_shapes: list<item: string>
  child 0, item: string
license: string
loss_scope: string
messages: list<item: struct<content: string, role: string, reasoning_content: string, target_component: string (... 2 chars omitted)
  child 0, item: struct<content: string, role: string, reasoning_content: string, target_component: string>
      child 0, content: string
      child 1, role: string
      child 2, reasoning_content: string
      child 3, target_component: string
native_input_ids: list<item: int64>
  child 0, item: int64
native_labels: list<item: int64>
  child 0, item: int64
native_model_family: string
native_model_revision: string
native_rendered_sha256: string
native_template_sha256: string
private_eval_content_used: bool
provenance: struct<benchmark_derived: bool, compaction_changed_semantics: bool, contrast_manifest_names: list<it (... 227 chars omitted)
  child 0, benchmark_derived: bool
  child 1, compaction_changed_semantics: bool
  child 2, contrast_manifest_names: list<item: null>
      child 0, item: null
  child 3, evaluator_content_accessed: bool
  child 4, license_class: string
  child 5, source_id: string
  child 6, source_kind: strin
...
tring
  child 1, schedule.native.json: struct<bytes: int64, sha256: string>
      child 0, bytes: int64
      child 1, sha256: string
  child 2, target.native.jsonl: struct<bytes: int64, sha256: string>
      child 0, bytes: int64
      child 1, sha256: string
updates: int64
replay_rows: int64
source_bindings: struct<audit.json: string, manifest.json: string, replay.jsonl: string, schedule.json: string, targe (... 16 chars omitted)
  child 0, audit.json: string
  child 1, manifest.json: string
  child 2, replay.jsonl: string
  child 3, schedule.json: string
  child 4, target.jsonl: string
schema: string
maximum_sequence_tokens: int64
counters: struct<action_cardinality: struct<0: int64, 1: int64, multi: int64>, availability: struct<irrelevant (... 215 chars omitted)
  child 0, action_cardinality: struct<0: int64, 1: int64, multi: int64>
      child 0, 0: int64
      child 1, 1: int64
      child 2, multi: int64
  child 1, availability: struct<irrelevant_menu_present: int64, relevant_menu_present: int64, tools_absent: int64>
      child 0, irrelevant_menu_present: int64
      child 1, relevant_menu_present: int64
      child 2, tools_absent: int64
  child 2, reasoning_mode: struct<off: int64, on: int64>
      child 0, off: int64
      child 1, on: int64
  child 3, system_condition: struct<generic_non_osaurus: int64, no_system: int64, raptor_current: int64>
      child 0, generic_non_osaurus: int64
      child 1, no_system: int64
      child 2, raptor_current: int64
rows: int64
to
{'artifacts': {'replay.native.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'schedule.native.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'target.native.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}}, 'assistant_tokens': Value('int64'), 'benchmark_content_used': Value('bool'), 'counters': {'action_cardinality': {'0': Value('int64'), '1': Value('int64'), 'multi': Value('int64')}, 'availability': {'irrelevant_menu_present': Value('int64'), 'relevant_menu_present': Value('int64'), 'tools_absent': Value('int64')}, 'reasoning_mode': {'off': Value('int64'), 'on': Value('int64')}, 'system_condition': {'generic_non_osaurus': Value('int64'), 'no_system': Value('int64'), 'raptor_current': Value('int64')}}, 'data_authority': Value('bool'), 'maximum_sequence_tokens': Value('int64'), 'model_bindings': {'chat_template.jinja': Value('string'), 'config.json': Value('string'), 'modeling_bailing_moe_v3.py': Value('string'), 'tokenizer_config.json': Value('string')}, 'private_eval_content_used': Value('bool'), 'replay_rows': Value('int64'), 'rows': Value('int64'), 'schema': Value('string'), 'source_bindings': {'audit.json': Value('string'), 'manifest.json': Value('string'), 'replay.jsonl': Value('string'), 'schedule.json': Value('string'), 'target.jsonl': Value('string')}, 'status': Value('string'), 'target_rows': Value('int64'), 'training_authority': Value('bool'), 'updates': Value('int64')}
because column names don't match
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 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              assistant_tokens: int64
              benchmark_content_used: bool
              capability: string
              closure_kind: string
              contamination_state: string
              contrast_group_id: string
              difficulty_band: string
              effective_tokens: int64
              effective_tokens_denominator: int64
              effective_tokens_numerator: int64
              execution_language: string
              id: string
              intent_class: string
              interface_shapes: list<item: string>
                child 0, item: string
              license: string
              loss_scope: string
              messages: list<item: struct<content: string, role: string, reasoning_content: string, target_component: string (... 2 chars omitted)
                child 0, item: struct<content: string, role: string, reasoning_content: string, target_component: string>
                    child 0, content: string
                    child 1, role: string
                    child 2, reasoning_content: string
                    child 3, target_component: string
              native_input_ids: list<item: int64>
                child 0, item: int64
              native_labels: list<item: int64>
                child 0, item: int64
              native_model_family: string
              native_model_revision: string
              native_rendered_sha256: string
              native_template_sha256: string
              private_eval_content_used: bool
              provenance: struct<benchmark_derived: bool, compaction_changed_semantics: bool, contrast_manifest_names: list<it (... 227 chars omitted)
                child 0, benchmark_derived: bool
                child 1, compaction_changed_semantics: bool
                child 2, contrast_manifest_names: list<item: null>
                    child 0, item: null
                child 3, evaluator_content_accessed: bool
                child 4, license_class: string
                child 5, source_id: string
                child 6, source_kind: strin
              ...
              tring
                child 1, schedule.native.json: struct<bytes: int64, sha256: string>
                    child 0, bytes: int64
                    child 1, sha256: string
                child 2, target.native.jsonl: struct<bytes: int64, sha256: string>
                    child 0, bytes: int64
                    child 1, sha256: string
              updates: int64
              replay_rows: int64
              source_bindings: struct<audit.json: string, manifest.json: string, replay.jsonl: string, schedule.json: string, targe (... 16 chars omitted)
                child 0, audit.json: string
                child 1, manifest.json: string
                child 2, replay.jsonl: string
                child 3, schedule.json: string
                child 4, target.jsonl: string
              schema: string
              maximum_sequence_tokens: int64
              counters: struct<action_cardinality: struct<0: int64, 1: int64, multi: int64>, availability: struct<irrelevant (... 215 chars omitted)
                child 0, action_cardinality: struct<0: int64, 1: int64, multi: int64>
                    child 0, 0: int64
                    child 1, 1: int64
                    child 2, multi: int64
                child 1, availability: struct<irrelevant_menu_present: int64, relevant_menu_present: int64, tools_absent: int64>
                    child 0, irrelevant_menu_present: int64
                    child 1, relevant_menu_present: int64
                    child 2, tools_absent: int64
                child 2, reasoning_mode: struct<off: int64, on: int64>
                    child 0, off: int64
                    child 1, on: int64
                child 3, system_condition: struct<generic_non_osaurus: int64, no_system: int64, raptor_current: int64>
                    child 0, generic_non_osaurus: int64
                    child 1, no_system: int64
                    child 2, raptor_current: int64
              rows: int64
              to
              {'artifacts': {'replay.native.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'schedule.native.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'target.native.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}}, 'assistant_tokens': Value('int64'), 'benchmark_content_used': Value('bool'), 'counters': {'action_cardinality': {'0': Value('int64'), '1': Value('int64'), 'multi': Value('int64')}, 'availability': {'irrelevant_menu_present': Value('int64'), 'relevant_menu_present': Value('int64'), 'tools_absent': Value('int64')}, 'reasoning_mode': {'off': Value('int64'), 'on': Value('int64')}, 'system_condition': {'generic_non_osaurus': Value('int64'), 'no_system': Value('int64'), 'raptor_current': Value('int64')}}, 'data_authority': Value('bool'), 'maximum_sequence_tokens': Value('int64'), 'model_bindings': {'chat_template.jinja': Value('string'), 'config.json': Value('string'), 'modeling_bailing_moe_v3.py': Value('string'), 'tokenizer_config.json': Value('string')}, 'private_eval_content_used': Value('bool'), 'replay_rows': Value('int64'), 'rows': Value('int64'), 'schema': Value('string'), 'source_bindings': {'audit.json': Value('string'), 'manifest.json': Value('string'), 'replay.jsonl': Value('string'), 'schedule.json': Value('string'), 'target.jsonl': Value('string')}, 'status': Value('string'), 'target_rows': Value('int64'), 'training_authority': Value('bool'), 'updates': Value('int64')}
              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.

Raptor 0.5 harness corpus

The SFT corpus used to produce Raptor 0.5 from inclusionAI/Ling-3.0-tiny. It is small on purpose: the goal was tool-surface familiarity for the Osaurus harness, not new capabilities.

File	Rows	What it is
`target.native.jsonl`	348	Supervised examples (the actual teaching signal)
`replay.native.jsonl`	802	Base-model replay rows mixed in for retention
`schedule.native.json`	—	Training schedule over the 1,150 rows
`manifest.json`	—	SHA-256 manifest of the artifacts

1,150 rows total, 230,034 assistant-supervised tokens, max sequence 4,096. Trained as a rank-4 LoRA (alpha 8) and merged into base weights at BF16 after 4 optimizer updates.

Row format

Rows are pre-rendered against the Ling-3.0-tiny chat template (revision b61f4338, template SHA-256 recorded per row) and carry both the readable and the tokenized form:

messages, tools — the conversation and the tool menu offered (may be empty)
native_input_ids, native_labels — tokenized sequence; labels are -100 outside supervision_spans (tool observations are always masked)
availability — tools_absent / relevant menu / irrelevant menu
system_condition — which system-prompt variant the row was rendered under
reasoning_mode — on / off, bound per row
executable_verification — for rows grounded in runnable projects: language, run command, expected-stdout hash, and PASS_EXECUTABLE status
provenance hashes for the rendered text, source profile, tool menu, and system prompt, plus the source commit of the generator

Composition

reasoning on / off: 575 / 575
tool availability: 168 tools-absent · 148 relevant menu · 32 irrelevant menu
system condition: 144 current harness prompt · 108 no system · 96 generic non-Osaurus prompt
expected actions per row: 200 zero-action · 18 single-call · 130 multi-call

Zero-action rows are deliberate: answering directly when no offered tool fits is trained, not assumed.

How rows were admitted

Every candidate row passed a mechanical gate before it entered the corpus: render through the real chat template, every tool call parses, every called tool exists in the row's own tools[], every argument validates against the real schema, reasoning tags structurally correct for the row's mode, tokenizer round-trip lossless on special tokens, near-duplicate scan, and a plain-text final answer present. Tool schemas and system prompts were extracted from the Osaurus host source, not written by hand.

Notes

The tool-call dialect is Ling's native XML-arg format (<arg_key> / <arg_value> inside <tool_call>), while the tool definitions the model reads are JSON. Parsers that expect JSON tool calls will not match.
The corpus was verified do-no-harm: the merged model's failure set on a 144-run defect sweep is identical to the base model's, and base benchmark scores are unchanged.

Osaurus · eric@osaurus.ai

Downloads last month: 21

Total file size:

17.2 MB

Raptor 0.5 harness corpus

Contents

Row format

Composition

How rows were admitted

Notes