Hugging Face's logo

Datasets:
timodonnell
/
helico-training-recipes

Tags:
biology
protein
training-data
reproducibility
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
datasets: list<item: struct<manifest: struct<contact_definition: struct<comparison: string, diagonal_supervise (... 1772 chars omitted)
  child 0, item: struct<manifest: struct<contact_definition: struct<comparison: string, diagonal_supervised: bool, di (... 1760 chars omitted)
      child 0, manifest: struct<contact_definition: struct<comparison: string, diagonal_supervised: bool, distance: string, e (... 1574 chars omitted)
          child 0, contact_definition: struct<comparison: string, diagonal_supervised: bool, distance: string, exclude_elements: list<item: (... 180 chars omitted)
              child 0, comparison: string
              child 1, diagonal_supervised: bool
              child 2, distance: string
              child 3, exclude_elements: list<item: string>
                  child 0, item: string
              child 4, states: struct<contact: int64, masked: int64, no_contact: int64>
                  child 0, contact: int64
                  child 1, masked: int64
                  child 2, no_contact: int64
              child 5, storage: string
              child 6, symmetric: bool
              child 7, threshold_angstrom: double
              child 8, unknown_geometry: string
              child 9, version: string
          child 1, files: list<item: struct<path: string, sha256: string, size: int64>>
              child 0, item: struct<path: string, sha256: string, size: int64>
                  child 0, path: string
                  child 1, sha256: st
...
 sa (... 22 chars omitted)
                  child 0, bioassembly_dir: string
                  child 1, index: string
                  child 2, pdb_entries: int64
                  child 3, pdb_ids: string
                  child 4, role: string
                  child 5, sampling_records: int64
              child 1, validation: struct<bioassembly_dir: string, index: string, pdb_entries: int64, pdb_ids: string, role: string, sa (... 22 chars omitted)
                  child 0, bioassembly_dir: string
                  child 1, index: string
                  child 2, pdb_entries: int64
                  child 3, pdb_ids: string
                  child 4, role: string
                  child 5, sampling_records: int64
              child 2, validation_pool: struct<bioassembly_dir: string, index: string, pdb_entries: int64, pdb_ids: string, role: string, sa (... 22 chars omitted)
                  child 0, bioassembly_dir: string
                  child 1, index: string
                  child 2, pdb_entries: int64
                  child 3, pdb_ids: string
                  child 4, role: string
                  child 5, sampling_records: int64
      child 1, manifest_path: string
      child 2, manifest_sha256: string
      child 3, manifest_size: int64
      child 4, repo_id: string
      child 5, revision: string
      child 6, train_split: string
      child 7, validation_split: string
      child 8, weight: double
lock_sha256: string
name: string
schema_version: int64
to
{'schema_version': Value('int64'), 'name': Value('string'), 'datasets': List({'repo_id': Value('string'), 'revision': Value('string'), 'manifest': Value('string'), 'train_split': Value('string'), 'validation_split': Value('string'), 'weight': Value('float64')})}
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
              datasets: list<item: struct<manifest: struct<contact_definition: struct<comparison: string, diagonal_supervise (... 1772 chars omitted)
                child 0, item: struct<manifest: struct<contact_definition: struct<comparison: string, diagonal_supervised: bool, di (... 1760 chars omitted)
                    child 0, manifest: struct<contact_definition: struct<comparison: string, diagonal_supervised: bool, distance: string, e (... 1574 chars omitted)
                        child 0, contact_definition: struct<comparison: string, diagonal_supervised: bool, distance: string, exclude_elements: list<item: (... 180 chars omitted)
                            child 0, comparison: string
                            child 1, diagonal_supervised: bool
                            child 2, distance: string
                            child 3, exclude_elements: list<item: string>
                                child 0, item: string
                            child 4, states: struct<contact: int64, masked: int64, no_contact: int64>
                                child 0, contact: int64
                                child 1, masked: int64
                                child 2, no_contact: int64
                            child 5, storage: string
                            child 6, symmetric: bool
                            child 7, threshold_angstrom: double
                            child 8, unknown_geometry: string
                            child 9, version: string
                        child 1, files: list<item: struct<path: string, sha256: string, size: int64>>
                            child 0, item: struct<path: string, sha256: string, size: int64>
                                child 0, path: string
                                child 1, sha256: st
              ...
               sa (... 22 chars omitted)
                                child 0, bioassembly_dir: string
                                child 1, index: string
                                child 2, pdb_entries: int64
                                child 3, pdb_ids: string
                                child 4, role: string
                                child 5, sampling_records: int64
                            child 1, validation: struct<bioassembly_dir: string, index: string, pdb_entries: int64, pdb_ids: string, role: string, sa (... 22 chars omitted)
                                child 0, bioassembly_dir: string
                                child 1, index: string
                                child 2, pdb_entries: int64
                                child 3, pdb_ids: string
                                child 4, role: string
                                child 5, sampling_records: int64
                            child 2, validation_pool: struct<bioassembly_dir: string, index: string, pdb_entries: int64, pdb_ids: string, role: string, sa (... 22 chars omitted)
                                child 0, bioassembly_dir: string
                                child 1, index: string
                                child 2, pdb_entries: int64
                                child 3, pdb_ids: string
                                child 4, role: string
                                child 5, sampling_records: int64
                    child 1, manifest_path: string
                    child 2, manifest_sha256: string
                    child 3, manifest_size: int64
                    child 4, repo_id: string
                    child 5, revision: string
                    child 6, train_split: string
                    child 7, validation_split: string
                    child 8, weight: double
              lock_sha256: string
              name: string
              schema_version: int64
              to
              {'schema_version': Value('int64'), 'name': Value('string'), 'datasets': List({'repo_id': Value('string'), 'revision': Value('string'), 'manifest': Value('string'), 'train_split': Value('string'), 'validation_split': Value('string'), 'weight': Value('float64')})}
              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.

Helico training recipes

Versioned dataset mixtures for Helico model training. Each recipe lists Hugging Face dataset repositories, exact revisions, split names and relative sampling weights. The accompanying lock embeds every manifest, source revision and file checksum. A new data mixture gets a new recipe/lock and a new training run.

Available recipes

Recipe Datasets Training entries Validation entries
protenix-v1 Experimental PDB structures, weight 1.0 167,997 384

This preserves the default Protenix v1 cutoff and validation subset, with the 8P7U/6G23 FoldBench overlap excluded from further fine-tuning. It does not claim that the public index is the exact original checkpoint-training manifest.

Run setup

Download a recipe from a pinned revision of this repository, then initialize a run with the installed Helico tooling:

helico-datasets init-run recipes/protenix-v1.json \
  --run-dir /data/runs/contact-v1 --cache-dir /data/helico-registry

This freezes data.lock.json in the run directory and stages checksum-verified inputs into a shared cache. It records local paths separately in data.paths.json. Full staging needs roughly 2.2 TB for archives plus extracted data. For metadata inspection only:

helico-datasets prepare recipes/protenix-v1.lock.json --metadata-only \
  --cache-dir /data/helico-registry --output /tmp/data.metadata.json

Model code consumes the returned dataset splits, weights and source roots. TrainConfig(data_lock=lock) records the full lock in checkpoints and W&B, and refuses resuming a trained checkpoint with changed data. The metadata release provides an upstream Protenix data-config adapter; the full-scale Helico feature adapter remains a separate integration step. No model training is launched here.

To add a dataset, publish its own versioned dataset.json, split membership, checksums, supervision contract and leakage audit. Then add its commit-pinned reference and weight to a new recipe. The registry supports separate sources; each training adapter must explicitly support their feature/label formats.

Planned architecture

Helico architecture

SVG · PDF

Blue-gray components retain the AF3 architecture. Green marks contact conditioning, the contact prediction head and absorbing-mask supervision. Amber marks the planned inference-time search. The continuous coordinate diffusion model remains in place. MSAs remain available alongside contacts. The full tree search is planned; it was not implemented by the small pilot.

AF3 architecture reference. The diagram is generated by figures/plot_contact_architecture.py in the Helico repository layout.
Downloads last month
35
Total file size:
615 kB