Skip to content

CUDA input gradients for Living Weights on Flash Next: a Gated DeltaNet layer, 4-bit products, routed experts, gated attention (shared), and the hyper-connections and pocket (family) - #895

Closed
ZackSample wants to merge 49 commits into
ashhart:mainfrom
ZackSample:lw-c-fnback
Closed

CUDA input gradients for Living Weights on Flash Next: a Gated DeltaNet layer, 4-bit products, routed experts, gated attention (shared), and the hyper-connections and pocket (family)#895
ZackSample wants to merge 49 commits into
ashhart:mainfrom
ZackSample:lw-c-fnback

Commits

  1. Commits on Oct 14, 2026

  2. Commits on Oct 15, 2026