Execution attempt

Sign in with GitHub
← Experiment E5 · How much of the change from an untouched model to the same model after an APT4 transplant and continued pretraining is attributable to the continued pretraining rather than to the tokenizer replacement, when the original-tokenizer model receives identical continued pretraining?

Execution attempt · A44 · planned

Freezes the digit-probe spec: items, exemplars, prompt formats and decoding settings drawn from E2's probes.

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 @ b4c55e6e7d26917e239ccbe2f11886c4d4a733a2

Reference checked 2026-09-14 13:10 UTC. Later commits, branches or plan changes do not retarget this attempt.

Command
python scripts/08_probe_digits.py --build-spec
Working directory
experiments/E05-control-arm
Configuration paths
None
Parameters
Seed 20260706; 120 addition, 90 subtraction, 50 comparison and 40 sorting items, allocated across digit counts by largest remainder; 4 exemplars per task from the easiest remaining items; formats bare|en, bare|pl, pl_space|pl; batch 24; answers cut at the first newline.
Environment
3.12.13; Python, numpy PCG64; 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.

  • probes.jsonl · evaluation data · M63 probes.jsonl · raw output · Download

Delivered events

  1. Registered

    #1

    Freezes the digit-probe spec: items, exemplars, prompt formats and decoding settings drawn from E2's probes.

    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.

    • probe_digits_spec.json · https://github.com/stw2/tokenizer-science-tax/blob/b4c55e6e7d26917e239ccbe2f11886c4d4a733a2/experiments/E05-control-arm/configs/probe_digits_spec.json · public · sha256 e11c98fc129e… · 5618 bytes · M150

    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.