14 lines (11 loc) · 423 Bytes

Dataset Expansion Plan

Near-term targets

100 records per cognition data family under data/
100 records per distillation family under datasets/distillation_*
50 tasks per benchmark family under benchmarks/

Distillation families

cleanroom mind-shape SFT
ARC-Core ontology reasoning tasks
Arc-RAR receipt/trace discipline tasks
critique/revision preference pairs
quantization retention probes