End-to-End Test-Time Training for Long Context
IntermediateArnuv Tandon, Karan Dalal et al.Dec 29arXiv
This paper shows how a language model can keep learning while you use it, so it handles very long inputs without slowing down.
#Test-Time Training#Meta-learning#Long-context language modeling