BrainForge/Ministral-3-8B-Base-2512

Lower Reasoning donor weight to 0.5 in TIES merge

Set apply_yarn_scaling: false for vLLM <= 0.20 YaRN magnitude

Preserve trained thinking token rows through TIES

Add BrainForge thinking tokens

Add MergeKit configuration

Add BrainForge chat template

Declare untied 8B embedding topology

Duplicate from mistralai/Ministral-3-8B-Base-2512