Search papers, labs, and topics across Lattice.
1
0
2
Forget expensive LLM-as-judge checks: Proxy-GRM learns transferable rubrics for vision-language reward models with a lightweight proxy, achieving SOTA results with 4x less data.
