Countdown

The 'Countdown' dataset is a benchmark used to evaluate the performance of models on numerical reasoning tasks, specifically in the context of solving arithmetic problems.

Papers using Countdown (8)

Discrete Tilt Matching2026Escaping the Verifier: Learning to Reason via Demonstrations2025HPO: Hysteretic Policy Optimization for Stable and Efficient Training under Sparse-Reward Regime2026TRE: Encouraging Exploration in the Trust Region2026How Does RL Post-training Induce Skill Composition? A Case Study on Countdown2025Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective2025RL in Name Only? Analyzing the Structural Assumptions in RL post-training for LLMs2025To Backtrack or Not to Backtrack: When Sequential Search Limits Model Reasoning2025