zils-lite-v9a

A body for the Zils d3-lite soup pool (Bittensor subnet 127): a fine-tune of Qwen/Qwen3.5-0.8B in the chassis layout, BF16, language weights trained, vision tower from the parent.

Start: qhoenix/zils-lite-decision-v9 (the d3-lite soup of the 2026-11-07 round, Apache-2.0). Uniform weight average of four fine-tunes of v9 (two full fine-tunes with soft cross-entropy + Brier and KL-anchor rows to v9, lr 4e-6, about 1.3 h each; two LoRA r32 fine-tunes, one epoch, lr 1e-4, KL anchor 0.1), then moved 1.25x along the average change from v9.

Training data: exam-format decision items written with the public Zils authoring recipe (scripts/author_exam.py) by Claude Opus and by a local open model, from public training splits only, plus the public Zils relay teacher answers. No validator exam items and no public seed items were used. This repository holds weights only; check the source datasets' terms before reuse.