QuasiMoTTo uses quasi-Monte Carlo to produce correlated yet marginally correct samples from language models, matching i.i.d. pass@k with 25-47% fewer samples on reasoning benchmarks and 50% fewer RL training steps.
Polychromic objectives for reinforcement learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
QuasiMoTTo: Quasi-Monte Carlo Test-Time Scaling
QuasiMoTTo uses quasi-Monte Carlo to produce correlated yet marginally correct samples from language models, matching i.i.d. pass@k with 25-47% fewer samples on reasoning benchmarks and 50% fewer RL training steps.