SLAYER PROTOCOL · ENGINEERING LOG
FABRYKA AI / RESEARCH ARCHIVE
Research archive.
The record as it was written.
Engineering logs, experiment records, plans and resources from the former Slayer website, kept with their original dates, names and findings.
This is a historical document from the former Slayer website. Original dates, names and findings are retained for context. It is not a statement of current capabilities, policy or roadmap. Interactive tools and submissions are not available in this snapshot.
The archived records were written in Polish and are shown in their original language.
Current research agenda → · Research notes →01 / LOGS
Engineering and experiment logs
Working notes written during model training: measurements, regressions, recipe investigations and the experiment log.
001 · 2026-06-13
Munin 1.0 nie zrobił CPT. Receptura wyciekła w nazwie katalogu
DZIENNIK BADAŃ / CZERWIEC 2026
Log eksperymentów
02 / PLANS AND NOTES
Plans and notes
Training plans, data proposals and status updates as they were published at the time.
propozycja · linia v3 · czysta z konstrukcji
v3 — propozycja miksu danych (czysta linia)
V4 plan · feedback open · po autopsji v3
V4 plan — autopsja regresji i bezpieczny harness
update · stan na 2026-06-21
Na czym stoimy — update
SOTA 2026 · public roadmap · feedback open
SOTA roadmap 2026 + komentarze
metody fine-tuningu · SOTA 2026
Trening — metody fine-tuningu (SOTA 2026)
01 · styl · v1 · przegląd ręczny
Styl: co base Qwen robi źle po polsku · Slayer
program badawczy · synthetic CPT
Wiedza — wstrzykiwanie polskiego długiego ogona (synthetic CPT)
03 / RESOURCES
Benchmarks and datasets
How results were measured, which benchmarks stayed closed and the catalogue of published datasets and models.
metodologia · piątka startowa
Benchmarki — metodologia
poza zakresem pomiaru
Benchmarki zamknięte
Dane · modele · pochodzenie