Rishab Balasubramanian, Pin-Jie Lin, Rituraj Sharma, Anjie Fang, Fardin Abdi
We investigate whether post-trained capabilities can be transferred across models without retraining, with a focus on transfer across different model scales. We propose the Master Key Hypothesis, which states that model …
HuggingFace
Sai Srinivas Kancheti, Aditya Kanade, Rohit Sinha, Vineeth N Balasubramanian, Tanuja Ganu
Multimodal reasoning models (MRMs) trained with reinforcement learning with verifiable rewards (RLVR) show improved accuracy on visual reasoning benchmarks. However, accuracy gains often come at the cost of reasoning …
HuggingFace
Chonghan Qin, Xiachong Feng, Weitao Ma, Xiaocheng Feng, Lingpeng Kong
Existing memory benchmarks for LLM agents evaluate explicit recall of facts, yet overlook implicit memory where experience becomes automated behavior without conscious retrieval. We introduce ImplicitMemBench, the first …
HuggingFace