Search papers, labs, and topics across Lattice.
2
0
4
0
LLM-as-a-Judge accuracy hinges on temperature settings, revealing a task-dependent sweet spot that defies the common practice of fixed values like 0.1 or 1.0.
Agent Skill frameworks can unlock surprisingly strong performance from moderately sized (12B-30B) language models in industrial settings, rivaling closed-source giants while boosting GPU efficiency.