ollama/mlxrunner/nn
Daniel Hiltgen cecac8bf1b
mlxrunner: add Inkling 2 Mini (#19799)
* mlxrunner: add Inkling 2 Mini

* mlx: fuse Inkling 2 Mini residual mixing on Metal

Fuse RMS normalization and query scoring in one kernel. Apply the
softmax weights in a second kernel to avoid storing per-query products.

* create: expose Inkling 2 Mini context length in model metadata

Copy model_max_length to max_position_embeddings during import so
ollama show reports the checkpoint's context length. Reject configs
that specify conflicting limits.

* mlxrunner: reduce Metal dispatches in Inkling 2 Mini decode

Reuse the prefill mask shader to compute relative position bias in one
dispatch for single-query attention. On M3 Ultra with Metal 3, compute
the loop controller's gain and forget projections in one matrix multiply.
2026-11-20 18:38:44 -08:00
..
conv.go mlxrunner: lay out nn one layer kind per file 2026-09-16 14:06:08 -07:00
embedding.go mlxrunner: lay out nn one layer kind per file 2026-09-16 14:06:08 -07:00
embedding_test.go mlxrunner: lay out nn one layer kind per file 2026-09-16 14:06:08 -07:00
linear.go mlx: support concurrent generation (#19197) 2026-11-10 13:24:12 -08:00
linear_test.go mlx: support concurrent generation (#19197) 2026-11-10 13:24:12 -08:00
multilinear.go mlxrunner: lay out nn one layer kind per file 2026-09-16 14:06:08 -07:00
norm.go mlxrunner: lay out nn one layer kind per file 2026-09-16 14:06:08 -07:00
norm_test.go mlx: support concurrent generation (#19197) 2026-11-10 13:24:12 -08:00
recurrent.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
recurrent_test.go Add Kev decision scoring to MLX (#18976) 2026-10-12 21:50:41 -07:00
rope.go mlx: support YaRN scaling in Qwen3 and Qwen3.5 (#19231) 2026-11-04 14:14:32 -08:00
rope_test.go mlx: support YaRN scaling in Qwen3 and Qwen3.5 (#19231) 2026-11-04 14:14:32 -08:00
sdpa.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
sdpa_test.go mlx: support concurrent generation (#19197) 2026-11-10 13:24:12 -08:00