Repository navigation

cuda gemma4: images on the 12B, and LayerNorm rows to 8192 wide (#1048 step 12) - #1183

Closed

jschmied wants to merge 23 commits into

ashhart:mainfrom

jschmied:g4-images-12b

Closed

cuda gemma4: images on the 12B, and LayerNorm rows to 8192 wide (#1048 step 12)#1183
jschmied wants to merge 23 commits into
ashhart:mainfrom
jschmied:g4-images-12b

Commits on Oct 18, 2026
- cuda norms: bf16 residual tails with a branch norm, a layer scalar and MLX's rounded rows
  jschmied
  committed
  88fd80f
Commits on Oct 19, 2026
Commits on Oct 20, 2026
- cuda window_attention: head dim 512 over a linear cache, for global attention
  jschmied
  committed
  7424800
Commits on Oct 21, 2026