zils-mini-s1

A body for the Zils d3-mini soup pool (Bittensor subnet 127): a fine-tune of Qwen/Qwen3.5-4B in the chassis layout, BF16, language weights trained, vision tower from the parent.

Start: the uniform weight average of four Apache-2.0 d3-mini house bodies: torchcast-ai/torchcast-decision-4b at 492bfb19, perplexity-ai/pplx-decider-flash-v1.2 at 9ff7cf98, vllm-sr/d3-mini at df99bf4c and Maincode/Matilda-Decision-v1-Mini at 39479a8a. Six independent one-epoch LoRA fine-tunes of that start (rank 32, alpha 64, lr 5e-5 to 1e-4, merged into the weights) were trained on 4.1k to 10k exam-format items with a cross-entropy + Brier loss; the body is the uniform average of the six merged weights.

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.