// ⚖ Alignment & Preference Optimization
RL with Verifiable Rewards (RLVR)
Rewards come from programmatic checks like math answers or unit tests.
可验证奖励强化学习 RLVR 2024 Advanced core
IN PLAIN WORDS
Instead of a learned reward model, check the final answer or run tests — rewards that cannot be flattered. It is key to o1/R1-style reasoner training; the term comes from Tulu 3.
Used in: Tulu 3, DeepSeek-R1, Qwen3 reasoning
TIMELINE · prerequisites → this → what builds on it
01Original paper
Tulu 3: Pushing Frontiers in Open Language Model Post-Training
Nathan Lambert, Jacob Morrison, Valentina Pyatkin, et al. · 2024
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
Then you can learn
Related:Process Reward Models (PRM)
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
On learning paths:Post-training & Reasoning Models
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
rlvr)