NanoFlux: Adversarial Dual-LLM Evaluation and Distillation For Multi-Domain Reasoning

Raviteja Anantha, Soheil Hor, Teodor Nicola Antoniu +1

We present NanoFlux, a novel adversarial framework for generating targeted training data to improve LLM reasoning, where adversarially-generated datasets containing fewer than 200…