Search papers, labs, and topics across Lattice.
Chinese Academy of Sciences
1
0
3
Context-augmented RL lets smaller MLLMs punch *way* above their weight, rivaling much larger models on reasoning tasks while dodging reward hacking.
