A d3-lite decision body (Qwen/Qwen3.5-0.8B family) for the Zils subnet 127 lite pool.
Fine-tuned from the house body vllm-sr/d3-lite (Apache-2.0) with LoRA (rank 16, AdamW lr 5e-5, 1 epoch) by distillation on train-split items of public datasets converted to the Zils decision format (mmlu, boolq, gsm8k, anli, glue, banking77, clinc150, When2Call, poem_sentiment, iSarcasm, sarcasm headlines, SVAMP, commonsense_qa, tweet_eval, 20newsgroups, glaive function calling). Targets are soft: 0.7 x the answer distribution of a seed average of our own mini fine-tunes (same recipe as malkolm777/zils-mini-decision-v5, Apache-2.0) + 0.3 x the gold label. Licensed under Apache-2.0; see LICENSE and NOTICE.
Re-committed as a new revision on 2026-11-14 (weights unchanged; retry of the same body).
model-00001-of-00001.safetensors 1.7 GBreadout.safetensors 1 MBvllm-sr/d3-lite