COUNTDOWN Dataset for ES vs GRPO Comparison
Dataset Description
Full Countdown dataset (2100 train + 100 test samples) for mathematical reasoning and arithmetic expression generation
Training split: 100.0% of available data
Training samples: 2,100
Validation samples: 0
Test samples: 100 (reserved for final evaluation)
Data Format
Each example contains:
