LoRA adapter fine-tuned on 10K human-written instruction-response pairs using Adaption's AutoScientist platform. Base model: GPT-OSS 120B.