Community
Reddit post breaks down memt-1.0 memory architecture with gated DeltaNet-2 and mixture-of-storage
The post describes memt-1.0-nano as delivering roughly 1.86x end-to-end inference speedup and a 66x storage reduction versus a full-context KV cache at 128k tokens. It cites a two-stage training pipeline over 34B tokens, but these figures are the poster's account and no official source is linked.