Execution attempt

Sign in with GitHub
← Experiment E9 · At a fixed 32k vocabulary, does reserving a small science slice of the tokenizer's training text remove most of the English-science fertility tax at under 5% cost to Polish fertility?

Execution attempt · A84 · planned

Builds the Polish, Python, mathematical web and general English training texts at pinned revisions, skipping the existing LaTeX text. The first 2,000 documents of the unshuffled Polish and English streams are skipped; this does not exclude E5's shuffled evaluation holdout, so one Polish holdout document is fully and 43 partly contained in pl.txt.

Succeeded · Registered by @stw2 via agent. Reported and received times are kept apart; no computation or result is verified.

Pinned source and configuration

https://github.com/stw2/tokenizer-science-tax @ a77445db2440ec00d756a3346b3bddedc69e85eb

Reference checked 2026-09-14 13:15 UTC. Later commits, branches or plan changes do not retarget this attempt.

Command
python scripts/01_build_train_corpus.py
Working directory
experiments/E09-vocabulary-allocation
Configuration paths
None
Parameters
quotas: pl 1,073,741,824, sci_python and sci_mathweb 71,582,788, en_general 214,748,365 bytes; first 2,000 documents skipped for pl and en_general; wikipedia.org hosts dropped for pl; Python files of 300 to 20,000 characters with ASCII ratio at least 0.9; no randomness
Environment
3.12.13; SentencePiece BPE trainer and Hugging Face tokenizers on CPU; no model weights; Apple silicon, 128 GB unified memory
Output directory
Not recorded

Inputs

Materials the registrant named when registering this attempt, by content identity; obtainability is derived from their location reports. Nothing is fetched or verified.

Delivered events

  1. Registered

    #1

    Builds the Polish, Python, mathematical web and general English training texts at pinned revisions, skipping the existing LaTeX text. The first 2,000 documents of the unshuffled Polish and English streams are skipped; this does not exclude E5's shuffled evaluation holdout, so one Polish holdout document is fully and 43 partly contained in pl.txt.

    reported · received · @stw2 via agent · posted to the Thread

  2. Started

    #2

    Started.

    reported · received · @stw2 via agent · attempt only

  3. Succeeded

    #3

    Completed with exit code 0.

    Exit code 0.

    • pl.txt · Not redistributed: ask the Room owner, or rebuild with scripts/01_build_train_corpus.py at the pinned revisions · restricted · sha256 5bd65b432c77… · 1078563622 bytes · M351
    • sci_python.txt · Not redistributed: ask the Room owner, or rebuild with scripts/01_build_train_corpus.py at the pinned revisions · restricted · sha256 779923c80658… · 73317005 bytes · M416
    • sci_mathweb.txt · Not redistributed: ask the Room owner, or rebuild with scripts/01_build_train_corpus.py at the pinned revisions · restricted · sha256 36f49b00f3d2… · 72118257 bytes · M417
    • en_general.txt · Not redistributed: ask the Room owner, or rebuild with scripts/01_build_train_corpus.py at the pinned revisions · restricted · sha256 fae778ee956e… · 216158541 bytes · M418
    • train_corpus_manifest.json · https://github.com/stw2/tokenizer-science-tax/blob/c5299916db7f792a209ff2944e66c673c3ebb817/experiments/E09-vocabulary-allocation/results/train_corpus_manifest.json · public · sha256 bc30c6ca5eae… · 2228 bytes · M419

    reported · received · @stw2 via agent · posted to the Thread

Report an event

The registrant’s capture tool normally delivers start and outcome events. Reporting here is the same author report with the browser as the reported time; it does not observe the process.

This attempt has a delivered outcome. A new execution is a new attempt.