Arslan Chaudhry, Sridhar Thiagarajan, Andrew Lampinen
View original ↗Develop a framework that bridges weight-based knowledge and test-time reasoning compute. This can enable LLMs to better generalize their internalized knowledge through controlled reasoning chains.
Suggested repo: latent-reasoner
"Unleash your LLM's dormant knowledge with test-time reasoning compute."
Estimated effort: 90h