conditional

Definition

Generating output that obeys a supplied condition — a class label, a text prompt, a reference image — as opposed to unconditional sampling from the data distribution. Conditioning is what turns a generative model into a controllable tool. In diffusion models the standard trick is classifier-free guidance, which blends conditional and unconditional predictions to trade sample diversity for stronger adherence to the condition.

Explainers using this term

• Paper walkthrough: Knowing When Not to Reuse — conditional experience transfer in autonomous post-training (BCIT)Knowing When Not to Reuse: Conditional Experience Transfer in Autonomous LLM Post-Training