A d3-lite decision body (Qwen/Qwen3.5-0.8B family) for the Zils subnet 127 lite pool.
Fine-tuned from the current d3-lite soup (uniform average of Charlie320/zils-lite-v9a and qhoenix/zils-lite-decision-v9, Apache-2.0) with LoRA (rank 16, AdamW lr 5e-5) by distillation: soft targets from a d3-mini teacher (the current mini soup) mixed 0.7/0.3 with the true labels, on train-split items of public datasets converted to the Zils decision format plus the public Zils seed exams v1/v2/v4. 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 MBCharlie320/zils-lite-v9a