mix30: builds and packs the training stream, builds the FVT transplant of Qwen2.5-0.5B, runs the gates, trains 1B tokens with checkpoint scoring, and scores the final checkpoint per document.

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 @ 1e3dd3f736856c404f72ee42eb26b08b2062c223

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

experiments/E06-math-code-rescue/scripts/00_bootstrap_cuda.sh

Command: bash scripts/00_bootstrap_cuda.sh mix30
Working directory: experiments/E06-math-code-rescue
Configuration paths: None
Parameters: config configs/e6_arms.json; arm mix30: shares of APT4 tokens Polish 0.56, English 0.14, OpenWebMath 0.15, Python code 0.15; base model Qwen2.5-0.5B; 1B tokens, sequence 2048, micro-batch 32 x gradient accumulation 8, learning rate 1e-4, 50 warmup steps, no decay, weight decay 0.1, clip 1.0, checkpoints every 100M tokens; stream seed 42, shuffle buffer 10,000, documents of at least 200 characters, pool headroom 1.03; scoring: first 200 documents per text, digit probe as frozen in configs/probe_digits_spec.json; original data-manifest directory 44191812 renamed to mix30
Environment: 3.10 (conda-based PyTorch container image); PyTorch CUDA, bf16, sdpa attention, gradient checkpointing, bitsandbytes 8-bit AdamW, Liger fused cross-entropy; HF_HUB_DISABLE_XET=1; 4090-class GPU, 24 GB
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.

Qwen2.5-0.5B · base model · M189 Qwen2.5-0.5B base model · model · Download
APT4 tokenizer · tokenizer · M46 Bielik-PL-11B-v3.0-Instruct · model · Ask the reporter
packed stream, 30% math and code · training data · M192 Packed recovery pretraining stream, 30% math and code · training data · Ask the reporter
Polish FineWeb2-HQ evaluation holdout · evaluation data · M129 pl_eval.jsonl · raw output · Download
English SlimPajama-6B evaluation holdout · evaluation data · M130 en_eval.jsonl · raw output · Ask the reporter
math_clean synthetic arithmetic set · evaluation data · M128 math_clean evaluation statements · dataset · Download
en-arxiv-abstracts.jsonl · evaluation data · M27 en-arxiv-abstracts.jsonl · raw output · Download
en-latex-methods.jsonl · evaluation data · M28 en-latex-methods.jsonl · raw output · Ask the reporter
en-python-code.jsonl · evaluation data · M29 en-python-code.jsonl · raw output · Ask the reporter
en-gsm8k.jsonl · evaluation data · M30 en-gsm8k.jsonl · raw output · Download
pl-science-pes.jsonl · evaluation data · M31 pl-science-pes.jsonl · raw output · Ask the reporter
pl-wiki-science.jsonl · evaluation data · M32 pl-wiki-science.jsonl · raw output · Download
pl-informal.jsonl · evaluation data · M33 pl-informal.jsonl · raw output · Ask the reporter
synthetic arithmetic probe items · probe items · M63 probes.jsonl · raw output · Download

Delivered events

Registered
#1
mix30: builds and packs the training stream, builds the FVT transplant of Qwen2.5-0.5B, runs the gates, trains 1B tokens with checkpoint scoring, and scores the final checkpoint per document.
reported 2026-07-08 20:02 UTC · received 2026-09-14 13:12 UTC · @stw2 via agent · posted to the Thread
Started
#2
Started.
reported 2026-07-08 20:02 UTC · received 2026-09-14 13:12 UTC · @stw2 via agent · attempt only
Succeeded
#3
Completed with exit code 0.
Exit code 0.
- data_manifests/mix30/raw_manifest.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/data_manifests/mix30/raw_manifest.json · public · sha256 c43c0f3b7f3a… · 1686 bytes · M248
- data_manifests/mix30/mix30.meta.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/data_manifests/mix30/mix30.meta.json · public · sha256 c09281321205… · 603 bytes · M249
- FVT transplant model.safetensors · Not kept: rebuilt deterministically by the init stage of scripts/00_bootstrap_cuda.sh · unavailable · sha256 246d363e3ae3… · M195
- env_manifest_mix30.txt · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/env_manifest_mix30.txt · public · sha256 14e643d6fe76… · 6158 bytes · M224
- bpb_mix30.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/bpb_mix30.jsonl · public · sha256 00fcaa4842f2… · 14227 bytes · M250
- probe_digits_mix30.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_mix30.jsonl · public · sha256 e4b15a0da85f… · 11031 bytes · M251
- probe_digits_raw/mix30/ckpt_00100M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00100M.jsonl · public · sha256 ede5ab3a9e53… · 244645 bytes · M252
- probe_digits_raw/mix30/ckpt_00200M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00200M.jsonl · public · sha256 a7db190b0ebb… · 243272 bytes · M253
- probe_digits_raw/mix30/ckpt_00300M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00300M.jsonl · public · sha256 7e28927c22e7… · 243733 bytes · M254
- probe_digits_raw/mix30/ckpt_00400M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00400M.jsonl · public · sha256 6904bd6adc20… · 244161 bytes · M255
- probe_digits_raw/mix30/ckpt_00500M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00500M.jsonl · public · sha256 6ba558ed7b82… · 244202 bytes · M256
- probe_digits_raw/mix30/ckpt_00600M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00600M.jsonl · public · sha256 21270ef9e4a3… · 244485 bytes · M257
- probe_digits_raw/mix30/ckpt_00700M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00700M.jsonl · public · sha256 db372914d76a… · 244586 bytes · M258
- probe_digits_raw/mix30/ckpt_00800M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00800M.jsonl · public · sha256 5d25b9814dfd… · 243100 bytes · M259
- probe_digits_raw/mix30/ckpt_00900M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_00900M.jsonl · public · sha256 564bfbd054fb… · 242301 bytes · M260
- probe_digits_raw/mix30/ckpt_01000M.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/ckpt_01000M.jsonl · public · sha256 83fb12bdd0f7… · 243973 bytes · M261
- probe_digits_raw/mix30/prompt_dump_ckpt_00100M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00100M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00200M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00200M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00300M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00300M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00400M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00400M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00500M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00500M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00600M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00600M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00700M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00700M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00800M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00800M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_00900M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_00900M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_ckpt_01000M.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_ckpt_01000M.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/prompt_dump_t0_base.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_t0_base.json · public · sha256 da2e51f25db4… · 777 bytes · M167
- probe_digits_raw/mix30/prompt_dump_t0_fvt.json · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/prompt_dump_t0_fvt.json · public · sha256 e7c16c523f52… · 859 bytes · M186
- probe_digits_raw/mix30/t0_base.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/t0_base.jsonl · public · sha256 026b934cde04… · 245310 bytes · M209
- probe_digits_raw/mix30/t0_fvt.jsonl · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/probe_digits_raw/mix30/t0_fvt.jsonl · public · sha256 6b2ed080fd56… · 264666 bytes · M210
- perdoc/mix30_final__pl.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__pl.npz · public · sha256 59521ec2d310… · 3702 bytes · M262
- perdoc/mix30_final__en.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__en.npz · public · sha256 53c5212d5202… · 3702 bytes · M263
- perdoc/mix30_final__sci_arxiv.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__sci_arxiv.npz · public · sha256 0c6ab54f2931… · 3702 bytes · M264
- perdoc/mix30_final__sci_latex.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__sci_latex.npz · public · sha256 570566d62b66… · 3702 bytes · M265
- perdoc/mix30_final__sci_python.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__sci_python.npz · public · sha256 38f6357a84fd… · 3702 bytes · M266
- perdoc/mix30_final__sci_gsm8k.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__sci_gsm8k.npz · public · sha256 21af9a41ae58… · 3702 bytes · M267
- perdoc/mix30_final__pl_wiki_sci.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__pl_wiki_sci.npz · public · sha256 c5b6fd3252a8… · 3702 bytes · M268
- perdoc/mix30_final__pl_pes.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__pl_pes.npz · public · sha256 50b62c4987db… · 3702 bytes · M269
- perdoc/mix30_final__pl_informal.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__pl_informal.npz · public · sha256 2321622d0d26… · 3702 bytes · M270
- perdoc/mix30_final__math_clean.npz · https://github.com/stw2/tokenizer-science-tax/blob/2457520fc798b19bee6893ac58e0357ec4c467d5/experiments/E06-math-code-rescue/results/perdoc/mix30_final__math_clean.npz · public · sha256 0aa977ccf47d… · 3702 bytes · M271
- checkpoint mix30/ckpt_01000M · Not redistributed: ask the Room owner, or rebuild with scripts/00_bootstrap_cuda.sh at the pinned revisions · restricted · sha256 782d085d90ce… · 776880358 bytes · M272
- checkpoints mix30/ckpt_00100M to ckpt_00900M · Not kept: deleted with the training machine; their scores are in results/bpb_mix30.jsonl and results/probe_digits_mix30.jsonl · unavailable
reported 2026-07-08 20:02 UTC · received 2026-09-14 13:12 UTC · @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.

Pinned source and configuration

Inputs

Delivered events

Registered

Started

Succeeded

Report an event