Hugging Face's logo

Datasets:
darkstar008
/
gmaps-parser-qwen3-4b-lora-data

Tasks:
Text Generation
Token Classification
Languages:
English
Size:
1K<n<10K
Tags:
air-traffic-control
Synthetic
faa-7110-65
asr-noise
runway-incursion
License:
Dataset card Files
xet
 Community
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:    CastError
Message:      Couldn't cast
SIMULATED: bool
TRAINING_ONLY: bool
text_only: bool
generator: string
generator_commit: string
phrasing_pools: list<item: string>
  child 0, item: string
heldout_pool_E_used: bool
seeds: struct<KLGA: list<item: int64>, KATL: list<item: int64>>
  child 0, KLGA: list<item: int64>
      child 0, item: int64
  child 1, KATL: list<item: int64>
      child 0, item: int64
counts: struct<KLGA: struct<scenarios: int64, transmissions: int64>, KATL: struct<scenarios: int64, transmis (... 14 chars omitted)
  child 0, KLGA: struct<scenarios: int64, transmissions: int64>
      child 0, scenarios: int64
      child 1, transmissions: int64
  child 1, KATL: struct<scenarios: int64, transmissions: int64>
      child 0, scenarios: int64
      child 1, transmissions: int64
transmissions_total: int64
instruction_counts: struct<CROSS: int64, CROSSING_COMPLETE: int64, GO_AROUND: int64, HOLD_SHORT: int64, LANDING_CLEARANC (... 75 chars omitted)
  child 0, CROSS: int64
  child 1, CROSSING_COMPLETE: int64
  child 2, GO_AROUND: int64
  child 3, HOLD_SHORT: int64
  child 4, LANDING_CLEARANCE: int64
  child 5, LUAW: int64
  child 6, OTHER: int64
  child 7, TAKEOFF_CLEARANCE: int64
  child 8, TAXI: int64
unique_texts: int64
excluded_for_heldout_overlap: struct<exact_overlap: int64, normalized_overlap: int64>
  child 0, exact_overlap: int64
  child 1, normalized_overlap: int64
overlap_with_heldout3_heldout4_after_exclusion: struct<exact: int64, normalized: int64>
  child 0, exact: int64
  child 1, normaliz
...
ild 1, groundable: int64
      child 2, ungroundable: int64
      child 3, ungroundable_reasons: int64
  child 4, harsh_reasons: struct<grounding: taxiway CY not heard: int64, grounding: taxiway Z not heard: int64, grounding: tax (... 451 chars omitted)
      child 0, grounding: taxiway CY not heard: int64
      child 1, grounding: taxiway Z not heard: int64
      child 2, grounding: taxiway F not heard: int64
      child 3, grounding: airline not heard and not resolved by the traffic list: int64
      child 4, grounding: taxiway Q not heard: int64
      child 5, grounding: taxiway G not heard: int64
      child 6, grounding: taxiway P not heard: int64
      child 7, grounding: taxiway U not heard: int64
      child 8, grounding: taxiway Y not heard: int64
      child 9, grounding: taxiway C not heard: int64
      child 10, grounding: taxiway F4 not heard: int64
      child 11, grounding: taxiway L3 not heard: int64
      child 12, grounding: taxiway L not heard: int64
  child 5, unprompted_reasons: struct<grounding: taxiway F not heard: int64, grounding: taxiway U not heard: int64, grounding: taxi (... 24 chars omitted)
      child 0, grounding: taxiway F not heard: int64
      child 1, grounding: taxiway U not heard: int64
      child 2, grounding: taxiway F4 not heard: int64
prompt_source: struct<KLGA: string, KATL: string>
  child 0, KLGA: string
  child 1, KATL: string
seed: int64
overlap_with_heldout_after_drop: int64
variants: list<item: string>
  child 0, item: string
to
{'SIMULATED': Value('bool'), 'TRAINING_ONLY': Value('bool'), 'text_only': Value('bool'), 'source_split': Value('string'), 'generator': Value('string'), 'generator_commit': Value('string'), 'seed': Value('int64'), 'variants': List(Value('string')), 'noise_model': {'file': Value('string'), 'learned_from': List(Value('string')), 'never_from': List(Value('string')), 'dev_pairs': {'unprompted': Value('int64'), 'prompted': Value('int64')}, 'p_digits': {'unprompted': {'flight': Value('float64'), 'runway': Value('float64'), 'taxiway_number': Value('float64'), 'other': Value('float64')}, 'prompted': {'flight': Value('float64'), 'runway': Value('float64'), 'taxiway_number': Value('float64'), 'other': Value('float64')}}}, 'gold_correction_W1': {'taxiways_inside_N_registration_removed': Value('int64'), 'note': Value('string')}, 'prompt_source': {'KLGA': Value('string'), 'KATL': Value('string')}, 'rows_written': Value('int64'), 'include_ungroundable': Value('bool'), 'per_variant': {'clean': {'rows': Value('int64'), 'groundable': Value('int64')}, 'unprompted': {'rows': Value('int64'), 'groundable': Value('int64'), 'ungroundable': Value('int64'), 'ungroundable_reasons': Value('int64')}, 'prompted': {'rows': Value('int64'), 'groundable': Value('int64')}, 'harsh': {'rows': Value('int64'), 'groundable': Value('int64'), 'ungroundable': Value('int64'), 'ungroundable_reasons': Value('int64')}, 'harsh_reasons': {'grounding: taxiway CY not heard': Value('int64'), 'grounding: taxiway Z not heard': Value('int64'), 'grounding: taxiway F not heard': Value('int64'), 'grounding: airline not heard and not resolved by the traffic list': Value('int64'), 'grounding: taxiway Q not heard': Value('int64'), 'grounding: taxiway G not heard': Value('int64'), 'grounding: taxiway P not heard': Value('int64'), 'grounding: taxiway U not heard': Value('int64'), 'grounding: taxiway Y not heard': Value('int64'), 'grounding: taxiway C not heard': Value('int64'), 'grounding: taxiway F4 not heard': Value('int64'), 'grounding: taxiway L3 not heard': Value('int64'), 'grounding: taxiway L not heard': Value('int64')}, 'unprompted_reasons': {'grounding: taxiway F not heard': Value('int64'), 'grounding: taxiway U not heard': Value('int64'), 'grounding: taxiway F4 not heard': Value('int64')}}, 'overlap_with_heldout_after_drop': Value('int64'), 'pairs_asr_jsonl': {'sha256': Value('string'), 'bytes': Value('int64'), 'tracked_in_git': Value('bool'), 'why': Value('string')}, 'script_sha256': Value('string')}
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 2951, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2461, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2486, 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 547, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 430, 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
              SIMULATED: bool
              TRAINING_ONLY: bool
              text_only: bool
              generator: string
              generator_commit: string
              phrasing_pools: list<item: string>
                child 0, item: string
              heldout_pool_E_used: bool
              seeds: struct<KLGA: list<item: int64>, KATL: list<item: int64>>
                child 0, KLGA: list<item: int64>
                    child 0, item: int64
                child 1, KATL: list<item: int64>
                    child 0, item: int64
              counts: struct<KLGA: struct<scenarios: int64, transmissions: int64>, KATL: struct<scenarios: int64, transmis (... 14 chars omitted)
                child 0, KLGA: struct<scenarios: int64, transmissions: int64>
                    child 0, scenarios: int64
                    child 1, transmissions: int64
                child 1, KATL: struct<scenarios: int64, transmissions: int64>
                    child 0, scenarios: int64
                    child 1, transmissions: int64
              transmissions_total: int64
              instruction_counts: struct<CROSS: int64, CROSSING_COMPLETE: int64, GO_AROUND: int64, HOLD_SHORT: int64, LANDING_CLEARANC (... 75 chars omitted)
                child 0, CROSS: int64
                child 1, CROSSING_COMPLETE: int64
                child 2, GO_AROUND: int64
                child 3, HOLD_SHORT: int64
                child 4, LANDING_CLEARANCE: int64
                child 5, LUAW: int64
                child 6, OTHER: int64
                child 7, TAKEOFF_CLEARANCE: int64
                child 8, TAXI: int64
              unique_texts: int64
              excluded_for_heldout_overlap: struct<exact_overlap: int64, normalized_overlap: int64>
                child 0, exact_overlap: int64
                child 1, normalized_overlap: int64
              overlap_with_heldout3_heldout4_after_exclusion: struct<exact: int64, normalized: int64>
                child 0, exact: int64
                child 1, normaliz
              ...
              ild 1, groundable: int64
                    child 2, ungroundable: int64
                    child 3, ungroundable_reasons: int64
                child 4, harsh_reasons: struct<grounding: taxiway CY not heard: int64, grounding: taxiway Z not heard: int64, grounding: tax (... 451 chars omitted)
                    child 0, grounding: taxiway CY not heard: int64
                    child 1, grounding: taxiway Z not heard: int64
                    child 2, grounding: taxiway F not heard: int64
                    child 3, grounding: airline not heard and not resolved by the traffic list: int64
                    child 4, grounding: taxiway Q not heard: int64
                    child 5, grounding: taxiway G not heard: int64
                    child 6, grounding: taxiway P not heard: int64
                    child 7, grounding: taxiway U not heard: int64
                    child 8, grounding: taxiway Y not heard: int64
                    child 9, grounding: taxiway C not heard: int64
                    child 10, grounding: taxiway F4 not heard: int64
                    child 11, grounding: taxiway L3 not heard: int64
                    child 12, grounding: taxiway L not heard: int64
                child 5, unprompted_reasons: struct<grounding: taxiway F not heard: int64, grounding: taxiway U not heard: int64, grounding: taxi (... 24 chars omitted)
                    child 0, grounding: taxiway F not heard: int64
                    child 1, grounding: taxiway U not heard: int64
                    child 2, grounding: taxiway F4 not heard: int64
              prompt_source: struct<KLGA: string, KATL: string>
                child 0, KLGA: string
                child 1, KATL: string
              seed: int64
              overlap_with_heldout_after_drop: int64
              variants: list<item: string>
                child 0, item: string
              to
              {'SIMULATED': Value('bool'), 'TRAINING_ONLY': Value('bool'), 'text_only': Value('bool'), 'source_split': Value('string'), 'generator': Value('string'), 'generator_commit': Value('string'), 'seed': Value('int64'), 'variants': List(Value('string')), 'noise_model': {'file': Value('string'), 'learned_from': List(Value('string')), 'never_from': List(Value('string')), 'dev_pairs': {'unprompted': Value('int64'), 'prompted': Value('int64')}, 'p_digits': {'unprompted': {'flight': Value('float64'), 'runway': Value('float64'), 'taxiway_number': Value('float64'), 'other': Value('float64')}, 'prompted': {'flight': Value('float64'), 'runway': Value('float64'), 'taxiway_number': Value('float64'), 'other': Value('float64')}}}, 'gold_correction_W1': {'taxiways_inside_N_registration_removed': Value('int64'), 'note': Value('string')}, 'prompt_source': {'KLGA': Value('string'), 'KATL': Value('string')}, 'rows_written': Value('int64'), 'include_ungroundable': Value('bool'), 'per_variant': {'clean': {'rows': Value('int64'), 'groundable': Value('int64')}, 'unprompted': {'rows': Value('int64'), 'groundable': Value('int64'), 'ungroundable': Value('int64'), 'ungroundable_reasons': Value('int64')}, 'prompted': {'rows': Value('int64'), 'groundable': Value('int64')}, 'harsh': {'rows': Value('int64'), 'groundable': Value('int64'), 'ungroundable': Value('int64'), 'ungroundable_reasons': Value('int64')}, 'harsh_reasons': {'grounding: taxiway CY not heard': Value('int64'), 'grounding: taxiway Z not heard': Value('int64'), 'grounding: taxiway F not heard': Value('int64'), 'grounding: airline not heard and not resolved by the traffic list': Value('int64'), 'grounding: taxiway Q not heard': Value('int64'), 'grounding: taxiway G not heard': Value('int64'), 'grounding: taxiway P not heard': Value('int64'), 'grounding: taxiway U not heard': Value('int64'), 'grounding: taxiway Y not heard': Value('int64'), 'grounding: taxiway C not heard': Value('int64'), 'grounding: taxiway F4 not heard': Value('int64'), 'grounding: taxiway L3 not heard': Value('int64'), 'grounding: taxiway L not heard': Value('int64')}, 'unprompted_reasons': {'grounding: taxiway F not heard': Value('int64'), 'grounding: taxiway U not heard': Value('int64'), 'grounding: taxiway F4 not heard': Value('int64')}}, 'overlap_with_heldout_after_drop': Value('int64'), 'pairs_asr_jsonl': {'sha256': Value('string'), 'bytes': Value('int64'), 'tracked_in_git': Value('bool'), 'why': Value('string')}, 'script_sha256': 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.

gmaps-parser-qwen3-4b-lora-data

Training transmissions by instruction

The scripted transmissions behind the training pairs, by instruction (from train/MANIFEST.json).

The data is synthetic: FAA JO 7110.65 phraseology on fictional callsigns and traffic, with zero overlap with the held-out sets (exact 0, normalized 0; train/MANIFEST.json).

Synthetic training and evaluation data for the GMAPS clearance parser (see the model card for darkstar008/gmaps-parser-qwen3-4b-lora). All of it is SIMULATED: fictional callsigns and traffic, phraseology templated from FAA JO 7110.65, runway and taxiway names from two public airport identifier lists (KLGA, KATL). No real ATC audio, no real transcripts, no audio files of any kind.

file rows what
train/pairs_asr.jsonl 7,065 chat-format pairs messages[system,user,assistant]; 4 variants per scripted transmission: clean, unprompted (Whisper-style noise), prompted (Whisper with an identifier prompt), harsh; fields id, src_id, airport, variant, groundable
train/ASR_NOISE_MANIFEST.json, train/MANIFEST.json generator commits, seeds, pools, overlap checks (0 overlap with every held-out set)
train/asr_noise_model.json word/number rewrite statistics learned from DEV splits only (Whisper large-v3-turbo transcripts of locally synthesized speech)
eval/heldout4_asr.jsonl 138 Whisper text of synthetic speech, gold from the script; NeMo Gym task format
eval/heldout4_asr_prompt.jsonl 138 same transmissions, Whisper run with an ICAO identifier prompt
eval/heldout5_script.jsonl 414 scripted text, generated after the parser freeze (blind set)

Gold taxiways use the callsign-aware rule (gold_rule: fixed): phonetic letters that belong to the callsign are not taxiways. The verifier (code/gym/resources_servers/gmaps_parser/app.py) scores record exact match, per-field match and groundable. Generate more with code/eval/synthetic/train_ft.py and code/train/augment_asr_noise.py.
Downloads last month
71
Total file size:
28.5 MB

Models trained or fine-tuned on darkstar008/gmaps-parser-qwen3-4b-lora-data