DeepSeek-R1-Distill-Qwen-7B

A self-supervised method called Reflective Recovery, which turns failed reasoning trajectories into recovery training data, raised accuracy on DeepSeek-R1-Distill-Qwen-7B from 30.0% to 37.5% on AIME 2…

A new arXiv paper (2608.28859v1) reports that DeepSeek-R1-Distill-Qwen-7B's chain of thought runs about twice as long as needed, and researchers internalized a causal 'halt vector' at layer 18 to cut …

Researchers propose Batch-wise Adaptive Pruning, a training-free method for batched inference in Large Reasoning Models (LRMs), achieving a 39.7 percentage point average accuracy improvement over prio…