TY  - RPRT
TI  - From Reasoning to Answer: Empirical, Attention-Based and Mechanistic Insights into Distilled DeepSeek R1 Models
AU  - Jue Zhang
AU  - Qingwei Lin
AU  - Saravan Rajmohan
AU  - Dongmei Zhang
PY  - 2025
UR  - https://arxiv.org/abs/2509.23676
ID  - 2509.23676
ER  -