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.
Asterism fusion corpus v1.0.0
A searchable, citable corpus of the fusion-plasma literature: 24,101 works from Nuclear Fusion and Plasma Physics and Controlled Fusion (1960 to 2026), with full text, 165,951 passages and 3,893 reconstructed tables for the 3,494 articles that carry a Creative Commons licence on the version of record. Every record traces to a DOI. No PDFs are distributed. Code and the MCP server: https://github.com/Plaetorius/asterism
Use
uv tool install git+https://github.com/Plaetorius/asterism
asterism-mcp download # these archives, checksum-verified, into ~/.asterism/
claude mcp add asterism -- asterism-mcp mcp
Files
| Archive | Contents |
|---|---|
fusion-corpus-db-v1.0.0.tar.zst |
catalog.sqlite (metadata, passages, FTS index), tables.sqlite, LICENSES.md, DATASHEET.md, MANIFEST.sha256 |
fusion-corpus-jsonl-v1.0.0.tar.zst |
works, citations (776,000 reference links), fulltext, passages, tables as JSONL (CC BY material only) |
fusion-corpus-nc-sa-v1.0.0.tar.zst |
the 3 CC BY-NC-SA 3.0 papers (full text, passages, table). Not for commercial use |
SHA256SUMS lists the archive checksums; MANIFEST.sha256 inside the db and JSONL archives lists every file.
Composition
- Metadata layer: 24,101 works (all Crossref records of the two journals); 23,439 classified as research articles. Sources: Crossref (metadata facts) and OpenAlex (CC0).
- Full-text layer: 3,494 articles, 92 % from 2020 or later: do not count trends on it; use the metadata layer with its coverage fields.
- Abstracts: released for 3,495 works whose article is CC licensed; withheld (null) for the other 18,195.
Licences
The selection, structure, cleaning and table reconstruction are offered under CC BY 4.0. Each article stays
under its own licence: 3,146 CC BY 4.0, 345 CC BY 3.0 and 3 CC BY-NC-SA 3.0. Every record carries its DOI and
licence URL, and full-text records an attribution string. Reuse requires attribution and a statement of changes
(see LICENSES.md in the archive). No endorsement by the authors, IOP Publishing or the IAEA is implied.
Collection
Metadata was harvested from the Crossref and OpenAlex APIs on 2026-10-02. Full texts were obtained from the publisher under an arrangement with the publisher; to build your own corpus, obtain PDFs or JATS under your publisher's text-and-data-mining terms or from open repositories, then ingest them locally with the code repository.
Known defects
- Table coverage is about 88 % of detected captions. Without ruling lines, adjacent columns can merge, and multi-line cells can appear as extra rows. Check values against the article.
- Equations are flattened to linear text; sub/superscripts are heuristic. Two-column reading order is usually right but figure captions and side boxes can interrupt paragraphs.
- Abstracts are present for only 3,495 of 24,101 works.
Intended use
Retrieval and question answering with citations, text mining of open fusion literature, table lookup (verify values), bibliometrics on the metadata layer. Not suitable: commercial use of the NC-SA subset; whole-field claims from the full-text layer alone; using reconstructed table values unchecked.
Evidence that it helps
Pre-registered, audited studies (same model, Claude Opus 5.5, in every setup; "web search" is the assistant's built-in search and fetch): exact facts from recent papers, corpus 93 % vs web search 74 % (129 questions); values printed in tables, 99 % vs 57 % (69 questions); papers public only after the model's training cutoff, 98 % vs 48 % (124 fact questions). On open-ended questions and broad evidence gathering there was no significant difference. Protocols and results: https://github.com/Plaetorius/asterism-evals
Privacy
Every released text value was scanned for the publisher's per-download stamps, cover-page residue, the
maintainer's personal details, local paths and IP addresses: 0 must-not-ship hits (see DATASHEET.md). The
articles' own author names, affiliations and correspondence addresses are published as printed in the articles.
Take-down and error reports: https://github.com/Plaetorius/asterism/issues
Citation
See CITATION.cff in the code repository (a Zenodo DOI is added at release).
- Downloads last month
- 47