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
id: string
parent: string
source: string
atoms: struct<numbers: list<item: int64>, positions: list<item: list<item: double>>, cell: list<item: list< (... 38 chars omitted)
child 0, numbers: list<item: int64>
child 0, item: int64
child 1, positions: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 2, cell: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 3, pbc: list<item: bool>
child 0, item: bool
energy: double
forces: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
stress: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
composition_group: string
audit: struct<raw_energy_eV: double, corrected_energy_eV: double, corrected_energy_per_atom_eV: double, str (... 48 chars omitted)
child 0, raw_energy_eV: double
child 1, corrected_energy_eV: double
child 2, corrected_energy_per_atom_eV: double
child 3, stress_source_kbar: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
split_manifest_sha256: string
counts: struct<train/mptrj: int64, train/salex: int64, val/mptrj: int64, val/salex: int64>
child 0, train/mptrj: int64
child 1, train/salex: int64
child 2, val/mptrj: int64
child 3, val/salex: int64
protocol: string
parent_cap: int64
files: struct<train.jsonl: string, val.jsonl: string>
...
ics: struct<train: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: (... 264 chars omitted)
child 0, train: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: int64, atom_qu (... 71 chars omitted)
child 0, frames: int64
child 1, unique_parents: int64
child 2, unique_compositions: int64
child 3, total_atoms: int64
child 4, atom_quantiles: struct<min: double, median: double, p95: double, max: double>
child 0, min: double
child 1, median: double
child 2, p95: double
child 3, max: double
child 1, val: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: int64, atom_qu (... 71 chars omitted)
child 0, frames: int64
child 1, unique_parents: int64
child 2, unique_compositions: int64
child 3, total_atoms: int64
child 4, atom_quantiles: struct<min: double, median: double, p95: double, max: double>
child 0, min: double
child 1, median: double
child 2, p95: double
child 3, max: double
energy_convention: string
corrections_applied: bool
overlap_checks: struct<all_eval_compositions_excluded: bool, all_eval_parent_ids_excluded: bool, train_dev_compositi (... 39 chars omitted)
child 0, all_eval_compositions_excluded: bool
child 1, all_eval_parent_ids_excluded: bool
child 2, train_dev_composition_disjoint: bool
child 3, unique_frames: bool
to
{'protocol': Value('string'), 'sources': {'mptrj': {'source': Value('string'), 'version': Value('string'), 'counts': {'train': Value('int64'), 'val': Value('int64')}, 'provenance': {'url': Value('string'), 'bytes': Value('int64'), 'md5': Value('string'), 'name': Value('string'), 'sha256': Value('string'), 'actual_md5': Value('string'), 'entire_source_downloaded': Value('bool'), 'publisher_checksum_verified': Value('bool')}, 'entire_source_scanned': Value('bool'), 'frames_scanned': Value('int64'), 'parents_scanned': Value('int64'), 'members': List(Value('null')), 'preparation_seconds': Value('float64'), 'parent_cap': Value('int64'), 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}}, 'salex': {'source': Value('string'), 'version': Value('string'), 'counts': {'train': Value('int64'), 'val': Value('int64')}, 'provenance': {'url': Value('string'), 'bytes': Value('int64'), 'name': Value('string'), 'sha256': Value('string'), 'actual_md5': Value('string'), 'entire_source_downloaded': Value('bool'), 'publisher_checksum_verified': Value('bool')}, 'entire_source_scanned': Value('bool'), 'frames_scanned': Value('int64'), 'parents_scanned': Value('int64'), 'members': List({'name': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'frames': Value('int64')}), 'preparation_seconds': Value('float64'), 'parent_cap': Value('int64'), 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}}}, 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}, 'counts': {'train/mptrj': Value('int64'), 'train/salex': Value('int64'), 'val/mptrj': Value('int64'), 'val/salex': Value('int64')}, 'statistics': {'train': {'frames': Value('int64'), 'unique_parents': Value('int64'), 'unique_compositions': Value('int64'), 'total_atoms': Value('int64'), 'atom_quantiles': {'min': Value('float64'), 'median': Value('float64'), 'p95': Value('float64'), 'max': Value('float64')}}, 'val': {'frames': Value('int64'), 'unique_parents': Value('int64'), 'unique_compositions': Value('int64'), 'total_atoms': Value('int64'), 'atom_quantiles': {'min': Value('float64'), 'median': Value('float64'), 'p95': Value('float64'), 'max': Value('float64')}}}, 'protected_compositions': Value('int64'), 'split_manifest_sha256': Value('string'), 'energy_convention': Value('string'), 'corrections_applied': Value('bool'), 'element_reference_resource': {'url': Value('string'), 'sha256': Value('string')}, 'overlap_checks': {'all_eval_compositions_excluded': Value('bool'), 'all_eval_parent_ids_excluded': Value('bool'), 'train_dev_composition_disjoint': Value('bool'), 'unique_frames': Value('bool')}, 'selection': Value('string'), 'limitations': List(Value('string')), 'parent_cap': 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
id: string
parent: string
source: string
atoms: struct<numbers: list<item: int64>, positions: list<item: list<item: double>>, cell: list<item: list< (... 38 chars omitted)
child 0, numbers: list<item: int64>
child 0, item: int64
child 1, positions: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 2, cell: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
child 3, pbc: list<item: bool>
child 0, item: bool
energy: double
forces: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
stress: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
composition_group: string
audit: struct<raw_energy_eV: double, corrected_energy_eV: double, corrected_energy_per_atom_eV: double, str (... 48 chars omitted)
child 0, raw_energy_eV: double
child 1, corrected_energy_eV: double
child 2, corrected_energy_per_atom_eV: double
child 3, stress_source_kbar: list<item: list<item: double>>
child 0, item: list<item: double>
child 0, item: double
split_manifest_sha256: string
counts: struct<train/mptrj: int64, train/salex: int64, val/mptrj: int64, val/salex: int64>
child 0, train/mptrj: int64
child 1, train/salex: int64
child 2, val/mptrj: int64
child 3, val/salex: int64
protocol: string
parent_cap: int64
files: struct<train.jsonl: string, val.jsonl: string>
...
ics: struct<train: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: (... 264 chars omitted)
child 0, train: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: int64, atom_qu (... 71 chars omitted)
child 0, frames: int64
child 1, unique_parents: int64
child 2, unique_compositions: int64
child 3, total_atoms: int64
child 4, atom_quantiles: struct<min: double, median: double, p95: double, max: double>
child 0, min: double
child 1, median: double
child 2, p95: double
child 3, max: double
child 1, val: struct<frames: int64, unique_parents: int64, unique_compositions: int64, total_atoms: int64, atom_qu (... 71 chars omitted)
child 0, frames: int64
child 1, unique_parents: int64
child 2, unique_compositions: int64
child 3, total_atoms: int64
child 4, atom_quantiles: struct<min: double, median: double, p95: double, max: double>
child 0, min: double
child 1, median: double
child 2, p95: double
child 3, max: double
energy_convention: string
corrections_applied: bool
overlap_checks: struct<all_eval_compositions_excluded: bool, all_eval_parent_ids_excluded: bool, train_dev_compositi (... 39 chars omitted)
child 0, all_eval_compositions_excluded: bool
child 1, all_eval_parent_ids_excluded: bool
child 2, train_dev_composition_disjoint: bool
child 3, unique_frames: bool
to
{'protocol': Value('string'), 'sources': {'mptrj': {'source': Value('string'), 'version': Value('string'), 'counts': {'train': Value('int64'), 'val': Value('int64')}, 'provenance': {'url': Value('string'), 'bytes': Value('int64'), 'md5': Value('string'), 'name': Value('string'), 'sha256': Value('string'), 'actual_md5': Value('string'), 'entire_source_downloaded': Value('bool'), 'publisher_checksum_verified': Value('bool')}, 'entire_source_scanned': Value('bool'), 'frames_scanned': Value('int64'), 'parents_scanned': Value('int64'), 'members': List(Value('null')), 'preparation_seconds': Value('float64'), 'parent_cap': Value('int64'), 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}}, 'salex': {'source': Value('string'), 'version': Value('string'), 'counts': {'train': Value('int64'), 'val': Value('int64')}, 'provenance': {'url': Value('string'), 'bytes': Value('int64'), 'name': Value('string'), 'sha256': Value('string'), 'actual_md5': Value('string'), 'entire_source_downloaded': Value('bool'), 'publisher_checksum_verified': Value('bool')}, 'entire_source_scanned': Value('bool'), 'frames_scanned': Value('int64'), 'parents_scanned': Value('int64'), 'members': List({'name': Value('string'), 'bytes': Value('int64'), 'sha256': Value('string'), 'frames': Value('int64')}), 'preparation_seconds': Value('float64'), 'parent_cap': Value('int64'), 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}}}, 'files': {'train.jsonl': Value('string'), 'val.jsonl': Value('string')}, 'counts': {'train/mptrj': Value('int64'), 'train/salex': Value('int64'), 'val/mptrj': Value('int64'), 'val/salex': Value('int64')}, 'statistics': {'train': {'frames': Value('int64'), 'unique_parents': Value('int64'), 'unique_compositions': Value('int64'), 'total_atoms': Value('int64'), 'atom_quantiles': {'min': Value('float64'), 'median': Value('float64'), 'p95': Value('float64'), 'max': Value('float64')}}, 'val': {'frames': Value('int64'), 'unique_parents': Value('int64'), 'unique_compositions': Value('int64'), 'total_atoms': Value('int64'), 'atom_quantiles': {'min': Value('float64'), 'median': Value('float64'), 'p95': Value('float64'), 'max': Value('float64')}}}, 'protected_compositions': Value('int64'), 'split_manifest_sha256': Value('string'), 'energy_convention': Value('string'), 'corrections_applied': Value('bool'), 'element_reference_resource': {'url': Value('string'), 'sha256': Value('string')}, 'overlap_checks': {'all_eval_compositions_excluded': Value('bool'), 'all_eval_parent_ids_excluded': Value('bool'), 'train_dev_composition_disjoint': Value('bool'), 'unique_frames': Value('bool')}, 'selection': Value('string'), 'limitations': List(Value('string')), 'parent_cap': Value('int64')}
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.
AI4Sci atomistic materials adaptation
Prepared inputs for draft task T0-RSI/ai4sci-tasks PR 49. This public release contains adaptation data, split metadata and measured baseline records. Evaluation geometries and targets are fetched from pinned upstream sources by the task materializer; they are not mirrored here. Download on the operator host, then mount only allowed files. Never pass the entire data directory to a model process.
| Component | Count | Use |
|---|---|---|
| MPTrj+sAlex training | 500,000 | Fit E/F/S potential; equal source quotas |
| Labeled development | 5,000 | Checkpoint selection; equal source quotas |
| Physical validation | 1,200 | WBM 1,000; MDR phonons 100; MP elasticity 100 |
| Physical test | 6,000 | WBM 5,000; MDR phonons 500; MP elasticity 500; operator only |
| Failure regression | 28 | Diagnostic inputs, not a representative score |
Full source releases were scanned with deterministic SHA-256 priority sampling, max four frames per source-parent, no atom cap. Training and labeled development are composition-disjoint. All 178,247 compositions and parent IDs in the original 279,043-case evaluation universe were excluded from adaptation. Protocol files preserve that original partition and the label-blind stratified compact panels. These exclusions do not establish independence from UMA pretraining.
Training JSONL rows contain IDs, periodic geometry, raw total energy (eV), forces (eV/angstrom), and ASE tensile-positive stress (eV/angstrom^3). MPTrj stresses are converted from compression-positive kbar; sAlex stresses retain their convention. No MP2020 correction is applied to training energies. Each prepared panel separates geometry-only inputs from reference targets. Test references must stay outside development and prediction mounts. Upstream labels are public; this is a runtime separation policy, not a claim that the source data are secret.
manifest.json records compressed and original checksums and expected paths.
The Git task pins the HF commit and manifest checksum, and downloads only explicit
components. Physical panel counts below describe the task, not mirrored HF payloads. All transformations, source file hashes, and split provenance are in
training/manifest.json ; evaluation source pins are in the task repository. Files are gzip JSONL
or JSON/YAML; use the task materializer rather than datasets.load_dataset.
records/standard-500k-v1 contains measured 3-epoch eight-H100 training results,
46,875-step learning curves, actual configs, and 1,200-case validation results.
The 6,000-case test was prepared but not evaluated. No UMA checkpoint or full raw
source archive is redistributed here. Obtain UMA separately with approved access.
This is a custom adaptation benchmark, not an official leaderboard submission.
See LICENSES.md for component-specific terms, sources, authors and modifications.
- Downloads last month
- 59