zils-lite-a14 (malkolm777)
A d3-lite decision body (Qwen/Qwen3.5-0.8B family) for the Zils subnet 127 lite soup pool.
A uniform average (0.5/0.5 of every weight) of two students of the pool soup, started from the house bodies vllm-sr/d3-lite and Matilda-Decision-v1-Edge. Each student: one LoRA pass (rank 64, AdamW lr 5e-5, 1 epoch, seed 553) distilled from a uniform average of two public lite bodies of this pool (VadymPikulyk/zils-lite-dh2 and qhoenix/zils-lite-decision-avg-abc, both Apache-2.0) on about 25k rows: public datasets in the pool decision format plus the public Zils seed exams v1/v2/v4; soft targets 0.5 x the teacher answer distribution + 0.5 x the gold label.
Training items: exam-style decision items authored and verified by an LLM from train-split rows of public datasets (mmlu, boolq, gsm8k, anli, glue, banking77, clinc150, When2Call, poem_sentiment, iSarcasm, sarcasm headlines, SVAMP, commonsense_qa, tweet_eval, 20newsgroups, glaive function calling, BIG-bench tasks) and rows of the same public datasets in the pool's decision format; targets are soft (0.5 x the teacher's answer distribution + 0.5 x the gold label) where a teacher is named. Licensed under Apache-2.0; see LICENSE and NOTICE.