The dataset viewer is not available for this split.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
group: string
source_root: string
archive: string
files: int64
raw_bytes: int64
compressed_bytes: int64
sha256: string
verified: bool
verification: string
verification_seconds: double
compression: string
members: list<item: struct<path: string, bytes: int64, sha256: string>>
child 0, item: struct<path: string, bytes: int64, sha256: string>
child 0, path: string
child 1, bytes: int64
child 2, sha256: string
seconds: double
to
{'archive': Value('string'), 'raw_bytes': Value('int64'), 'compressed_bytes': Value('int64'), 'sha256': Value('string'), 'verified': Value('bool'), 'seconds': Value('float64'), 'members': List({'path': Value('string'), 'bytes': 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 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
group: string
source_root: string
archive: string
files: int64
raw_bytes: int64
compressed_bytes: int64
sha256: string
verified: bool
verification: string
verification_seconds: double
compression: string
members: list<item: struct<path: string, bytes: int64, sha256: string>>
child 0, item: struct<path: string, bytes: int64, sha256: string>
child 0, path: string
child 1, bytes: int64
child 2, sha256: string
seconds: double
to
{'archive': Value('string'), 'raw_bytes': Value('int64'), 'compressed_bytes': Value('int64'), 'sha256': Value('string'), 'verified': Value('bool'), 'seconds': Value('float64'), 'members': List({'path': Value('string'), 'bytes': 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.
DVL paper structures
Lossless structure archives prepared on 2026-10-05. These contain stored Jacobian sketches, VJP structures, matrix-free representations, and vocabulary readouts used in the DVL project. They are numerical research artifacts rather than a tabular dataset.
The 10 archives total 59,890,049,555 compressed bytes (59.89 GB), from 70,559,382,040 source bytes.
| Archive | Compressed GB | Contents |
|---|---|---|
| A31_all67.tar.zst | 13.435 | Historical Sketch-J cache for all 67 available models, including the paper selected48 subset and auxiliary caches. |
| B_vjp.tar.zst | 11.299 | Concept VJP directions and pullbacks; excludes intervention execution caches. |
| C_semantic_vjp.tar.zst | 0.750 | Semantic VJP structures. |
| C_matrixfree_hybrid.tar.zst | 11.371 | Matrix-free hybrid structures (basis_q and q_times_j). |
| FD_dense_small5.tar.zst | 3.551 | Dense vocabulary Jacobians for five smaller models. |
| FD_dense_large4.tar.zst | 5.407 | Dense vocabulary Jacobians for four larger models. |
| FD_vocab_small5.tar.zst | 5.102 | Derived vocabulary readouts for five smaller models. |
| FD_vocab_large4.tar.zst | 8.783 | Derived vocabulary readouts for four larger models. |
| A100_local_matrixfree3.tar.zst | 0.192 | Three historical A100 matrix-free blocks preserved as distinct variants. |
| DVL_storage_audit.tar.zst | 0.000 | Paper provenance audit, expected hashes, selected48 coverage, and restore instructions. |
Completeness
The available historical A31 structures cover all 48 paper-selected models. Sixteen of the 21 expected paper B_vjp pullback files were found and matched their recorded SHA-256 hashes. Five A100_2 pullbacks are unavailable in this release: Llama-3.2-1B-Instruct, Qwen2.5-0.5B-Instruct, Qwen2.5-1.5B-Instruct, Qwen2.5-3B-Instruct, Qwen3-1.7B. Historical matrix-free blocks are preserved separately and do not replace those missing pullbacks.
Integrity and restoration
All archives passed full Zstandard decompression/checksum validation. SHA256SUMS records the compressed archive hashes. Per-file paths, sizes, and source SHA-256 hashes are embedded under __backup_manifest__/.
sha256sum -c SHA256SUMS
mkdir -p restored_metric
zstd -dc A31_all67.tar.zst | tar -xf - -C restored_metric
Remote archives preserve paths relative to the original metric directory. The audit archive contains provenance reports and detailed restoration notes. Source model weights are not included.
- Downloads last month
- 25