common-pile-prefix-continuation

Chat-template-ready English prefix continuation examples derived from selected Common Pile components. This upload contains only judge-accepted rows from a larger generated Common Pile export.

Contents

• Format: gzip-compressed JSON Lines under data/*.jsonl.gz
• Schema: {"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}
• Files: 19
• Rows: 619,911
• Task: English prefix continuation

Sources

Accepted rows by source family:

Source family	Accepted rows
common-pile/arxiv_abstracts_filtered	79,693
common-pile/arxiv_papers_filtered	288,234
common-pile/library_of_congress	251,984

Effective source files:

Source file	Accepted rows
common-pile/arxiv_abstracts_filtered/arxiv-abstracts-dolma-0000.parquet	39,776
common-pile/arxiv_abstracts_filtered/arxiv-abstracts-dolma-0001.parquet	39,917
common-pile/arxiv_papers_filtered/arxiv-papers-0000.parquet	36,502
common-pile/arxiv_papers_filtered/arxiv-papers-0001.parquet	36,172
common-pile/arxiv_papers_filtered/arxiv-papers-0002.parquet	36,201
common-pile/arxiv_papers_filtered/arxiv-papers-0003.parquet	36,045
common-pile/arxiv_papers_filtered/arxiv-papers-0004.parquet	35,887
common-pile/arxiv_papers_filtered/arxiv-papers-0005.parquet	35,981
common-pile/arxiv_papers_filtered/arxiv-papers-0006.parquet	35,891
common-pile/arxiv_papers_filtered/arxiv-papers-0007.parquet	35,555
common-pile/library_of_congress/data/00000_loc_books.parquet	29,822
common-pile/library_of_congress/data/00001_loc_books.parquet	31,750
common-pile/library_of_congress/data/00002_loc_books.parquet	29,643
common-pile/library_of_congress/data/00003_loc_books.parquet	28,041
common-pile/library_of_congress/data/00004_loc_books.parquet	29,546
common-pile/library_of_congress/data/00005_loc_books.parquet	28,087
common-pile/library_of_congress/data/00006_loc_books.parquet	29,490
common-pile/library_of_congress/data/00007_loc_books.parquet	30,571
common-pile/library_of_congress/data/00008_loc_books.parquet	15,034

Example

{
  "messages": [
    {
      "role": "user",
      "content": "Continue the text naturally. A fully differential calculation in perturbative quantum chromodynamics is presented for the production of massive photon pairs at hadron colliders. All next-to-leading order perturbative contributions from quark-antiquark, gluon-(anti)quark, and gluon-gluon subprocesses are included, as well as all-orders resummation of initial-state gluon radiation valid at next-to-next-to-leading logarithmic accuracy. The region of phase space is specified in which the calculation is most reliable. Good agreement is demonstrated with data from the Fermilab Tevatron, and predictions are made for more detailed tests"
    },
    {
      "role": "assistant",
      "content": "with CDF and DO data. Predictions are shown for distributions of diphoton pairs produced at the energy of the Large Hadron Collider (LHC). Distributions of the diphoton pairs from the decay of a Higgs boson are contrasted with those produced from QCD processes at the LHC, showing that enhanced sensitivity to the signal can be obtained with judicious selection of events."
    }
  ]
}

Filtering

Rows were judged with google/gemma-4-31B-it.

Outcome	Rows
Audited	740,695
Accepted/uploaded	619,911
Rejected	120,784

Main rejection reasons: incoherent-or-ocr, url-or-reference-dump, metadata-boilerplate, empty-or-too-short. Machine-readable counts are in audit_summary.json.

The larger generated export had 19,043,384 rows; this upload keeps only rows with an explicit accepted audit decision.

Recreate

recreate_dataset.py can regenerate a similar dataset from local Parquet files:

python recreate_dataset.py \
  --input-root /path/to/source-parquet-root \
  --output-root data

The script also includes audit and filter subcommands for rerunning the judge/filter workflow with an OpenAI-compatible endpoint.