LEM/seeds
Snider 7bea00a401 feat: LEK-1 kernel A/B test — 29 models, P100 validation, curriculum pipeline
Full v2 scorer benchmark data across 29 models (20 base + 9 LEK-tuned):
- P20 (21 probes): All 29 models, 3 conditions each
- P100 (101 probes): Top 5 models + LEK-4B, publication-quality data

Key findings:
- LEK-1B (21.74) beats base 4B/12B/27B at P100 scale — no kernel needed
- Emergent realignment resistance: LEK models degrade with runtime kernel
- Gemma3-12B + JSON kernel = 23.66 (best kernel-boosted score)
- Family lineages: Mistral 3.80→14.58, Qwen regressed then recovered

New scripts: ab_test.py (v2 scorer), self_distill.py (curriculum generation),
extract_training.py, rephrase_probes.py, Phase 0/1 runners

New seeds: P01-P100 merged (101 probes), 404 rephrased variants,
50 creative prompts for Phase 0 baseline lock

27B curriculum design: 4-phase staged training targeting 25+ baseline

Co-Authored-By: Virgil <virgil@lethean.io>
2026-02-19 11:32:26 +00:00
..
expansions Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
regional Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-africa-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-cn-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-de-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-en-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-eu-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-me-all-seeds.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
lem-prompts.jsonl Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00
P01-P20.json LEM+LEK 2026-02-12 04:05:28 +00:00
P01-P100-rephrased.json feat: LEK-1 kernel A/B test — 29 models, P100 validation, curriculum pipeline 2026-02-19 11:32:26 +00:00
P01-P100.json feat: LEK-1 kernel A/B test — 29 models, P100 validation, curriculum pipeline 2026-02-19 11:32:26 +00:00
P21-P40.json LEM+LEK 2026-02-12 04:05:28 +00:00
P41-P60.json LEM+LEK 2026-02-12 04:05:28 +00:00
P61-P80.json LEM+LEK 2026-02-12 04:05:28 +00:00
P81-P100.json LEM+LEK 2026-02-12 04:05:28 +00:00
phase0-creative.json feat: LEK-1 kernel A/B test — 29 models, P100 validation, curriculum pipeline 2026-02-19 11:32:26 +00:00
test-200-english.json Add regional seeds, expansion rounds, scripts, HF cards, benchmark summary 2026-02-13 13:39:08 +00:00