Daniel Hiltgen 2cba7756c5 Gemma4 on MLX (#15244) ...

* gemma4: implement Gemma 4 model for MLX (text-only runtime)

* gemma4: two MoE + SWA prefill perf fixes

Two performance optimizations in the gemma4 forward pass

1. Memoize the sliding-window prefill mask across layers.
2. Softmax only over the selected experts in Router.Forward.

* review comments

2026-04-13 16:36:51 -07:00

..

create_test.go

mlx: mixed-precision quant and capability detection improvements (#15409)

2026-04-13 11:43:07 -07:00

create.go

Gemma4 on MLX (#15244)

2026-04-13 16:36:51 -07:00

quantize.go

mlx: add op wrappers for Conv2d, Pad, activations, trig, and masked SDPA (#14913)

2026-04-13 11:43:24 -07:00