Load Qwen3.5 adapters with the pinned model's attention shape - #959

Merged

bradhilton merged 5 commits into

mainfrom

dalinar/qwen35-pinned-config

Sep 25, 2026

Commits on Sep 25, 2026

Load Qwen3.5 adapters with the pinned model's attention shape

bradhilton
and
claude
committed
43adc88
Fill adapter dimensions only from a complete model shape, on both load paths

bradhilton
and
claude
committed
4c4b65d
Cover missing head size and empty revisions; pass the provider in the harness

bradhilton
and
claude
committed
e2264fb
Resolve each missing Qwen3.5 attention dimension from the pinned config

bradhilton
and
claude
committed
b7c3915
Exercise the old wrong head-size fallback in the heads-only test

bradhilton
and
claude
committed
a23028d