The dataset viewer is not available for this subset.
Exception: SplitsNotFoundError
Message: The split names could not be parsed from the dataset config.
Traceback: Traceback (most recent call last):
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
for split_generator in builder._split_generators(
~~~~~~~~~~~~~~~~~~~~~~~~~^
StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/webdataset/webdataset.py", line 80, in _split_generators
raise ValueError(
...<2 lines>...
)
ValueError: The TAR archives of the dataset should be in WebDataset format, but the files in the archive don't share the same prefix or the same types.
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
for split in get_dataset_split_names(
~~~~~~~~~~~~~~~~~~~~~~~^
path=dataset,
^^^^^^^^^^^^^
config_name=config,
^^^^^^^^^^^^^^^^^^^
token=hf_token,
^^^^^^^^^^^^^^^
)
^
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
info = get_dataset_config_info(
path,
...<6 lines>...
**config_kwargs,
)
File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.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.
TopG source nets and benchmark snapshot
The inputs, not the outputs. This is the provenance half of the
topg-geometric-frameworks
library: the generated nets that every framework in that library was decorated
from, plus the frozen benchmark snapshot its evaluation depends on.
It exists because neither was otherwise recoverable. The decorated frameworks are published; the source nets were not, and they cannot be regenerated. The TopG generator is a neural model served from a Gradio Space, so re-querying it would produce a different catalogue — the specific nets used to build the library exist only here. Without this dataset the library is usable but its provenance chain is broken: you can read the results and you cannot rebuild them from source.
Contents
| file | contents |
|---|---|
topg_nets_full6.tar.gz |
18,987 generated nets — skeletons.jsonl + the cifs/ they reference |
topg_nets_full7.tar.gz |
20,219 generated nets, same structure |
topg_nets_full.tar.gz |
the earlier generation campaign (batched query output + analysis) |
mp_benchmark.tar.gz |
the Materials Project snapshot and the frozen benchmark split |
metric_expanded_aux.tar.gz |
uid_offsets.json (byte offsets into the metric-expanded library) and propose_index_slim.json.gz |
The nets
Each record in skeletons.jsonl carries id, kind (1m/2m/3m), the
requested coordination query, measured_sg vs topg_sg (what was asked for
vs what the generator delivered — these disagree more often than one would
like, and that disagreement is itself a documented finding), n_atoms,
times_drawn, and a relative path into the accompanying cifs/ directory.
The CIFs are therefore required, not optional: cif is a filename, not
structure text.
The benchmark snapshot
mp_benchmark.tar.gz contains bench_split.json, the frozen tune/holdout
partition. This matters more than its size suggests: the project's central
methodological rule is that the holdout is evaluated once, at the end, and that
rule is only enforceable if the split is preserved byte-for-byte. It also
contains the Materials Project structure dump the split indexes into — kept
because re-downloading from the MP API later returns a different snapshot,
which would silently break comparability with every frozen benchmark number.
Relationship to the library
Reproduction path:
topg_nets_full{,6,7} -> Voronoi decoration -> topg-geometric-frameworks
(the published library)
mp_benchmark -> evaluation of the proposer built on that library
uid values here and in topg-geometric-frameworks belong to the same
generation and are joinable. They are not comparable with the older
topg-decoration-library release, whose namespace shares zero overlap; a join
across those two returns nothing, silently.
Attribution
The mp_benchmark contents derive from the Materials Project
(https://materialsproject.org), used under CC-BY-4.0 — please cite them if you
use that portion. Everything else is CC-BY-4.0 from this work.
Citation
A paper describing the library and the crystal-chemical results derived from it is in preparation.
- Downloads last month
- 62