cleanup: drop abandoned 350M config and v1-only data prep

remove generic docs; keep run journal

remove docs/A100_KICKOFF_PROMPT_V2.md

remove docs/A100_KICKOFF_PROMPT.md

hf-conversion: ship MatildaForCausalLM via trust_remote_code + journal

eval: token-matched comparison vs Pythia-160M (10 ckpts x 5 tasks)

instance run 2026-05-30: liger 0.8.0 workarounds + journal

v1.5 pivot: 152M (18L x 768d) hero config, ReLU2 FFN, final logit soft-cap. 350M kept as reference.

v2 prep: keep_last=10 for 125GB disk, v2 kickoff prompt, strip cost language

v2: 363M hero run (Muon hybrid, WSD, Liger, SmolLM 75/15/10 mix)

initial commit