Execution attempt

Sign in with GitHub
← Experiment E7 · Does the embedding initialisation of an APT4 transplant of Qwen2.5-1.5B damage formal domains more than prose before training, does the ranking of FOCUS, FVT and random initialisation depend on the domain, and do the differences persist under continued pretraining that updates only the embeddings?

Execution attempt · A74 · planned

Scores the 18 stage-2 checkpoints in bits per byte on ten domains with whitespace-gated tokenizer loads.

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 @ 80716b075735d909dacf90838f0e4ced6af46be8

Reference checked 2026-09-14 13:12 UTC. Later commits, branches or plan changes do not retarget this attempt.

Command
python scripts/03_score_bpb.py --model s2_fvt_ckpt_00030M=ckpts_s2/fvt/ckpt_00030M s2_fvt_ckpt_00060M=ckpts_s2/fvt/ckpt_00060M s2_fvt_ckpt_00090M=ckpts_s2/fvt/ckpt_00090M s2_fvt_ckpt_00100M=ckpts_s2/fvt/ckpt_00100M s2_fvt_ckpt_00120M=ckpts_s2/fvt/ckpt_00120M s2_fvt_ckpt_00150M=ckpts_s2/fvt/ckpt_00150M s2_focus_ckpt_00030M=ckpts_s2/focus/ckpt_00030M s2_focus_ckpt_00060M=ckpts_s2/focus/ckpt_00060M s2_focus_ckpt_00090M=ckpts_s2/focus/ckpt_00090M s2_focus_ckpt_00100M=ckpts_s2/focus/ckpt_00100M s2_focus_ckpt_00120M=ckpts_s2/focus/ckpt_00120M s2_focus_ckpt_00150M=ckpts_s2/focus/ckpt_00150M s2_random_ckpt_00030M=ckpts_s2/random/ckpt_00030M s2_random_ckpt_00060M=ckpts_s2/random/ckpt_00060M s2_random_ckpt_00090M=ckpts_s2/random/ckpt_00090M s2_random_ckpt_00100M=ckpts_s2/random/ckpt_00100M s2_random_ckpt_00120M=ckpts_s2/random/ckpt_00120M s2_random_ckpt_00150M=ckpts_s2/random/ckpt_00150M --out results/bpb_e7s2.jsonl --perdoc results/perdoc_s2
Working directory
experiments/E07-embedding-init
Configuration paths
None
Parameters
18 invocations with one --model spec each, in the order of argv, as checkpoints arrived; max-docs 200; seq 2048 tokens; add_special_tokens false; bf16; tokenizers through load_tokenizer(), whose fallback to the pristine APT4 tokenizer was logged for 18 of 18 checkpoints; perdoc_s2/ digest is the directory digest of its 180 files
Environment
3.12.13; PyTorch MPS, bf16, row-wise scoring; 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

    Scores the 18 stage-2 checkpoints in bits per byte on ten domains with whitespace-gated tokenizer loads.

    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.

    • bpb_e7s2.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/7d450d1f62209ff5ac5415543443d2a25db74b7e/experiments/E07-embedding-init/results/bpb_e7s2.jsonl · public · sha256 14ccbd2a528a… · 21973 bytes · M389
    • bpb_e7s2.jsonl as produced · Archived private repository, experiments/E7/results/bpb_e7s2.jsonl at commit 1fac185; the public copy replaces 18 absolute model paths with paths relative to the experiment folder · unavailable · sha256 dd988de0b828… · 22963 bytes · M390
    • perdoc_s2/ · https://github.com/stw2/tokenizer-science-tax/tree/7d450d1f62209ff5ac5415543443d2a25db74b7e/experiments/E07-embedding-init/results/perdoc_s2 · public · sha256 7c0499ade064… · 666360 bytes · M391

    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.