DOT Collection The models of our paper "Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning Models" accepted by ACL2026 • 6 items • Updated about 11 hours ago