zils-flash-fj3
A body for the Zils d3-flash soup pool (Bittensor subnet 127): a fine-tune of Qwen/Qwen3.5-9B in the chassis layout, BF16, language weights trained, vision tower from the parent.
Fine-tune of the d3-flash house model: the start is vllm-sr/d3-flash (Apache-2.0, built on Qwen/Qwen3.5-9B); credit for the start belongs to its authors. Procedure: three LoRA (rank 32, merged) fine-tunes of that start with different seeds, one shuffled pass each over the 30,000 rows below at learning rate 5e-5, were averaged uniformly with equal weights. Training targets were soft and label-free: the answer probabilities of the current d3-mini soup average (teacher distillation from the d3-mini pool's soup: the uniform weight average of Zayaan/zils-mini-hm4, qhoenix/zils-mini-decision-t2avg and joamCriff123/zils-mini-m1113b).
Training data (30,000 rows): rows of the train split of tasksource/tasksource-jev-typed-decisions (about 630 source tasks; sources with commercial or unspecified licence metadata only, non-commercial sources excluded; sources whose names match BIG-bench tasks dropped; no Glaive rows), used only as unlabeled distillation inputs (their labels were not used). No exam-format written items, no validator exam items, no public seed exam items and no customer data were used. This repository holds weights only; check the source datasets' terms before reuse.