1 paper · 1 filter
Amrit Diggavi Seshadri
With the size and cost of large transformer-based language models growing, recently, there has been interest in shortcut casting of early transformer hidden-representations to fina…