shaban2024/Qwen3.5-2B-BOS-Q8-NonThinking

Model card v0.5 (joint-23, bench v0.5, usage, limitations)

v0.5: joint-23 (eval 0.937, gate 6W-2T) — jasan dobitak nad v0.4

v0.4: joint-02 merge (bez kolapsa u non-thinking)

v0.3: joint replay merge (tool 0.917, safety 1.0 nt)

v0.2: think-alignment merge (tool 0.833, arg 0.9-1.0)

v0.1: AgentMujo Q8 (merged LoRA, bench tool_selection 0.833)

Update README.md

Update README.md

initial commit