Tutorials

Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room πŸ™‚

Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room πŸ™‚ Gated DeltaNet-2 is here. πŸš€ πŸ”₯ New paper: Gated DeltaNet-2: Decoupling Erase and Write in Linea

DGX agentx-post
tutorialsjeremy-howard--x

Gated DeltaNet-2 is almost exactly RWKV-7's DPLR recurrence, not acknowledging the elephant in the room πŸ™‚ Gated DeltaNet-2 is here. πŸš€ πŸ”₯ 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. πŸ† πŸ’‘ Here's the idea behind it: Linear attention squeezes …

Source: Jeremy Howard (X) | 2026-05-22

Loading related sources…