sGPO: Trading Inference FLOPs for Training Efficiency in RLVR
About
Standard Reinforcement Learning with Verifiable Rewards (RLVR) training allocates a fixed rollout budget to every query, without regard for what each query's difficulty means for the current policy. This leads to two symmetric failure modes: easy queries produce near-zero advantage because the policy already solves them, while unsolvable queries produce no signal because the policy never solves them. Both regimes waste training FLOPs without contributing to a learning gradient. We introduce sorted Group Policy Optimization (sGPO), a compute-efficient strategy that trades a small budget of inference FLOPs for a large reduction in wasted training FLOPs. The key insight is that cheap inference compute can serve as a single offline proxy for query difficulty. By generating a small batch of parallel samples per query under the initial policy, we obtain a model-aware empirical success rate. This motivates setting the training rollout group size to the inverse of this success rate, a practical rule that maximizes sample efficiency by extracting the most advantage per generated rollout. This single profiling pass simultaneously drives data filtering (removing trivial queries and sub-sampling unsolvable ones), adaptive group size allocation, and curriculum construction (scheduling queries from easy to hard). sGPO matches or exceeds baseline performance while reducing total training compute by a factor of three, with the upfront inference profiling cost included.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | HMMT 2025 | -- | 241 | |
| Scientific Question Answering | SciKnowEval | Accuracy (Biology)27.8 | 15 | |
| Mathematical Reasoning | AIME 2024 | Avg@16 Accuracy31 | 10 | |
| Mathematical Reasoning | AIME 2026 | AIME avg@16 Accuracy15.4 | 10 | |
| Mathematical Reasoning | HMMT 2026 | Accuracy@166.6 | 10 | |
| Mathematical Reasoning | AIME 2025 | Accuracy@1613.5 | 10 |