yangxw/countdown-backtracking

Step Back to Leap Forward: Self-Backtracking for Boosting Reasoning of Language Models train: 500K test (Seen Targets): 5k test (New Targets): 5k github: https://github.com/LAMDASZ-ML/Self-BackTracking