Tutorials
Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room π
Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room π Gated DeltaNet-2 is here. π π₯ New paper: Gated DeltaNet-2: Decoupling Erase and Write in Linea
Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room π Gated DeltaNet-2 is here. π π₯ New paper: Gated DeltaNet-2: Decoupling Erase and Write in Linear Attention Gated DeltaNet-2 outperforms KDA and Mamba-3, the latest and best recurrent architectures, head to head at 1.3B. π π‘ Here's the idea behind it: Linear attention squeezes β¦
Source: Jeremy Howard (X) | 2026-05-22