This model was produced by first merging Qwen/Qwen3-8B-Base with Qwen/Qwen3-8B, OpenDataArena/Qwen3-8B-ODA-Math-460k, mlabonne/Qwen3-8B-abliterated using the DELLA MergeKit method (della). AIM was then applied to that merged parent using calibration examples from HuggingFaceFW/fineweb-edu (sample-10BT), allenai/WildChat, open-web-math/open-web-math, allenai/wildjailbreak (train).
