X · @teortaxesTex
· X / Twitter
RT DailyPapers: Meta AI released Sparse Delta Memory SDM scales linear RNN hidden states to massive capacity through sparse addressing. It matches the…
RT DailyPapersMeta AI released Sparse Delta MemorySDM scales linear RNN hidden states to massive capacity through sparse addressing.It matches the recall of large context KV caches while maintaining the FLOP efficiency of Gated DeltaNet, boosting in-context learning and reasoning.