Colosso-owned merged model for SN47 EvolAI miner bootstrap.

This model was produced from the local SBE LoRA adapter evolai-qwen25-15b-lora-acceptance-20260605T103709 merged into Qwen/Qwen2.5-1.5B.

Training summary: 512 rows, 9000 steps, first loss 3.8946, last loss 0.2176, quality gate pass.