Built av_sft / ar_sft / rl parquets (+ .nla_meta.yaml sidecars) with raw activation vectors, used to warm-start and RL-train the PoC model.
Part of the NLA ordered-features (nested-dropout RL) PoC: an Activation
Verbalizer trained so its ~10 next-token-prediction features are ordered
most->least important, via random feature-prefix truncation (nested dropout)
during GRPO RL. Code: feat/nla-ordered-features-rl on… See the full description on the dataset page:
https://huggingface.co/datasets/syvb/nla-ordered-features-poc-data.
