RLVR
Reinforcement learning with verifiable rewards (RLVR) trains a model with rewards from an automatic checker, such as a correct maths answer or passing unit tests, instead of a learned reward model.
The term comes from the Tülu 3 post-training recipe. DeepSeek-R1 showed that large-scale RL on verifiable maths and code rewards can produce long chain-of-thought reasoning.
Verifiable rewards are hard to game in the way learned reward models are, but they only exist for tasks with checkable answers.
Learn more on Kojen
Related terms
- RLHFReinforcement learning from human feedback (RLHF) fine-tunes a model to maximise a reward model trained on human preference comparisons, usually with PPO and a penalty for drifting from the original model..
- DPODirect Preference Optimization (DPO) trains a model on pairs of preferred and rejected responses with a simple classification-style loss, matching RLHF's objective without a separate reward model or reinforcement learning..
- Agent evalAn agent eval measures an AI agent on multi-step tasks, scoring both the final outcome and how it got there, including tool calls, turns, cost and reliability over repeated runs..
Sources
Put the definitions to work.
Kojen evaluates, tunes and red-teams your agents with these methods built in: Wilson intervals, pass^k, LLM judges with visible components. Free to start.
Start training free