AMC-23

The 'AMC23' dataset/benchmark is used to evaluate reinforcement learning models' ability to perform multi-step reasoning while managing the trade-off between efficiency and accuracy in their responses.

Papers using AMC-23 (11)

Nemotron-CrossThink: Scaling Self-Learning beyond Math Reasoning2025 · 24 citesSqueeze the Soaked Sponge: Efficient Off-policy Reinforcement Finetuning for Large Language Model2025 · 10 citesThink Dense, Not Long: Dynamic Decoupled Conditional Advantage for Efficient Reasoning2026Beyond Variance: Prompt-Efficient RLVR via Rare-Event Amplification and Bidirectional Pairing2026Long Chain-of-Thought Compression via Fine-Grained Group Policy Optimization2026Masked-and-Reordered Self-Supervision for Reinforcement Learning from Verifiable Rewards2025Confidence Is All You Need: Few-Shot RL Fine-Tuning of Language Models2025Reinforcement Learning for Reasoning in Small LLMs: What Works and What Doesn't2025Qwen2.5-Math Technical Report: Toward Mathematical Expert Model via Self-Improvement2024 · 12 cites$\texttt{SPECS}$: Faster Test-Time Scaling through Speculative Drafts2025Stop Summation: Min-Form Credit Assignment Is All Process Reward Model Needs for Reasoning2025