Inference Time Scaling
A technique for improving output quality by allocating more computation during the inference stage while keeping model weights unchanged. Approaches include generating long chains of thought, sampling multiple candidates and voting, reranking with verifiers, and tree search. It shifts additional compute to generation time rather than training time.
Timeline (last 90 days)
The course pipeline covers, in order: small model fundamentals, Verifier, Inference Time Scaling, GRPO reinforcement learning, distillation.
In the reasoning model pipeline, the verifier is used in the inference-time scaling stage to score multiple sampled answers and pick the best one, provides reward signals in GRPO training, and is used in the distillation stage to filter high-quality reasoning trajectories.
All Facts (2)
The course pipeline covers, in order: small model fundamentals, Verifier, Inference Time Scaling, GRPO reinforcement learning, distillation.
50%UnverifiedIn the reasoning model pipeline, the verifier is used in the inference-time scaling stage to score multiple sampled answers and pick the best one, provides reward signals in GRPO training, and is used in the distillation stage to filter high-quality reasoning trajectories.
50%