Remove failed RFT1 adapter (trained on RS data only, missing curated dataset)

#1

RFT1 scored 41.1% compile rate vs R5's 68.6% β€” catastrophic regression.

Root cause: trained on 14,951 rejection-sampled pairs only, without the 3,430 curated R5 pairs. Violated the accumulate-and-retrain rule.

Removing to prevent anyone from using a broken adapter.

the-clanker-lover changed pull request status to merged

Sign up or log in to comment