The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
original_path: string
remote_path: string
previous_remote_path: string
group: string
bytes: int64
sha256: string
git_blob_sha1: string
source_audits: list<item: string>
child 0, item: string
reasons: list<item: string>
child 0, item: string
domain: string
dataset: string
variant: string
run_id: string
hyperparameters: struct<model: string, configured_max_steps: int64, batch_per_device: int64, devices: int64, accumula (... 436 chars omitted)
child 0, model: string
child 1, configured_max_steps: int64
child 2, batch_per_device: int64
child 3, devices: int64
child 4, accumulation: int64
child 5, global_batch_size: int64
child 6, context: int64
child 7, learning_rate: double
child 8, seed: int64
child 9, hac_threshold: int64
child 10, data_mode: string
child 11, warmup_steps: int64
child 12, weight_decay: double
child 13, precision: string
child 14, gradient_clip_val: double
child 15, model_config_overrides: struct<attn_pdrop: double, activation_function: string>
child 0, attn_pdrop: double
child 1, activation_function: string
child 16, num_nodes: int64
child 17, configured_num_train: int64
child 18, scaling_subset_size: int64
child 19, scheduler_decay_steps: int64
child 20, scheduler_min_factor: double
paper_commit: string
original_bytes: int64
schema: string
created_pt: string
code_commit: string
excluded: list<item: struct<original_path: string, reason: string, evidence: string>>
child 0, item: struct<original_path: string, reason: string, evidence: string>
child 0, original_path: string
child 1, reason: string
child 2, evidence: string
repositories: struct<data: string, checkpoints: string, results: string>
child 0, data: string
child 1, checkpoints: string
child 2, results: string
source_payloads_unchanged: bool
legacy_paths_policy: string
category: string
original_files: int64
to
{'schema': Value('string'), 'paper_commit': Value('string'), 'code_commit': Value('string'), 'created_pt': Value('string'), 'excluded': List({'original_path': Value('string'), 'reason': Value('string'), 'evidence': Value('string')}), 'repositories': {'data': Value('string'), 'checkpoints': Value('string'), 'results': Value('string')}, 'category': Value('string'), 'original_files': Value('int64'), 'original_bytes': Value('int64'), 'source_payloads_unchanged': Value('bool'), 'legacy_paths_policy': 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 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
original_path: string
remote_path: string
previous_remote_path: string
group: string
bytes: int64
sha256: string
git_blob_sha1: string
source_audits: list<item: string>
child 0, item: string
reasons: list<item: string>
child 0, item: string
domain: string
dataset: string
variant: string
run_id: string
hyperparameters: struct<model: string, configured_max_steps: int64, batch_per_device: int64, devices: int64, accumula (... 436 chars omitted)
child 0, model: string
child 1, configured_max_steps: int64
child 2, batch_per_device: int64
child 3, devices: int64
child 4, accumulation: int64
child 5, global_batch_size: int64
child 6, context: int64
child 7, learning_rate: double
child 8, seed: int64
child 9, hac_threshold: int64
child 10, data_mode: string
child 11, warmup_steps: int64
child 12, weight_decay: double
child 13, precision: string
child 14, gradient_clip_val: double
child 15, model_config_overrides: struct<attn_pdrop: double, activation_function: string>
child 0, attn_pdrop: double
child 1, activation_function: string
child 16, num_nodes: int64
child 17, configured_num_train: int64
child 18, scaling_subset_size: int64
child 19, scheduler_decay_steps: int64
child 20, scheduler_min_factor: double
paper_commit: string
original_bytes: int64
schema: string
created_pt: string
code_commit: string
excluded: list<item: struct<original_path: string, reason: string, evidence: string>>
child 0, item: struct<original_path: string, reason: string, evidence: string>
child 0, original_path: string
child 1, reason: string
child 2, evidence: string
repositories: struct<data: string, checkpoints: string, results: string>
child 0, data: string
child 1, checkpoints: string
child 2, results: string
source_payloads_unchanged: bool
legacy_paths_policy: string
category: string
original_files: int64
to
{'schema': Value('string'), 'paper_commit': Value('string'), 'code_commit': Value('string'), 'created_pt': Value('string'), 'excluded': List({'original_path': Value('string'), 'reason': Value('string'), 'evidence': Value('string')}), 'repositories': {'data': Value('string'), 'checkpoints': Value('string'), 'results': Value('string')}, 'category': Value('string'), 'original_files': Value('int64'), 'original_bytes': Value('int64'), 'source_payloads_unchanged': Value('bool'), 'legacy_paths_policy': Value('string')}
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.
MOSAIC NeurIPS 2026 — checkpoints
Paper-scoped scientific backup, organized as:
molecular/<dataset>/<variant>/...
non-molecular/<dataset>/<variant>/...
metadata/...
Datasets: GuacaMol (guacamol), MOSES (moses), COCONUT (coconut2), GEOM-Drugs disjoint split (geom-drugs), Macrocycle (macrocycle), Proteins (proteins), Planar (planar), SBM (sbm), SSBM (ssbm), and RSBM R=4 original/connected populations (rsbm-r4-original, rsbm-r4-connected).
Variants HAC-R, HAC-FR, MC-R, MC-FR, PH-R, and PH-FR denote the paper's HDT variants. SENT, HSENT-HAC-R, HSENT-MC-R, and HDTC-R retain their distinct meanings. Shared data/references are stored once under shared; HAC, MC, PH, HDTC, and Louvain in the data repository identify a reusable cache family, not an additional trained model.
Names and settings
Checkpoint names follow:
<dataset>_<variant>_<backbone>_steps<N>_batch<B>_ctx<L>_lr<rate>[_tau<T>]_seed<S>[_online|_precomputed]_last.ckpt
For example:
molecular/guacamol/HAC-FR/guacamol_HAC-FR_gpt2-xs_steps100000_batch256_ctx512_lr0.0006_tau4_seed42_last.ckpt
batch is the effective global batch including devices and gradient accumulation. tau is the HAC community-size threshold and is omitted for methods that do not use HAC. seed in the run name is the training seed; evaluation seeds are separate subdirectories. Full settings and exact paper locations appear in metadata/neurips2026_run_catalog.json; frozen training configurations are in the results repository alongside each run. Different context lengths, thresholds, schedules and model sizes remain distinguishable.
Data paths also identify the cached population, split/shard, coarsening family and threshold where applicable. Results are grouped by run, evaluation protocol, motif and evaluation seed. Cross-dataset figure/source metadata lives under metadata/.
Scope and integrity
This backup follows manuscript revision 559f541889ed6156d5af3e801055f25bc74cc3d1. It is not an archive of every exploratory experiment. Only LAST checkpoints are included; optimizer and training state are retained. Conditional results are motif-conditioned only. Historical HAC-FR-Top PGD files accidentally included in the previous selection are explicitly excluded because that variant is absent from this paper revision.
All scientific payloads are preserved byte-for-byte. Standardized names apply to repository paths. Frozen configurations, checkpoint fields and provenance may still contain original experiment labels or paths; changing those would alter the research artifacts. The manifest maps each standard path back to its exact original path and records its SHA256. Historical Git commits are retained; the previous layout is removed from the current repository tree after verification.
Restore
Download neurips2026_backup_index.json from the results repository and restore_neurips_backup.py from any of the three repositories. Install huggingface_hub, then run:
python restore_neurips_backup.py --index neurips2026_backup_index.json --destination /path/to/backup
The default produces <destination>/<data|checkpoints|results>/<standard repository path>. To reconstruct the original code-compatible project tree, add --layout original. Original absolute paths inside some frozen manifests may require an equivalent project-root alias. Use --category data|checkpoints|results and repeatable --path <standard-or-original-path> to select a subset. Downloads are revision-pinned and checksum-verified; identical existing files are skipped and different existing files are never overwritten.
Repositories: data, checkpoints, results.
- Downloads last month
- 71