ollama/mlx
Daniel Hiltgen 2bd4d6aabd
mlx: fuse Inkling residual mixing on Metal (#19808)
Compute the normalized residuals, query scores, and weighted sum in one
dispatch. Keep intermediate values in the threadgroup and preserve the
existing reduction order and rounding.
2026-11-20 20:05:50 -08:00
..
compat/mlx-c mlx: reject unsupported CUDA devices (#19521) 2026-11-04 09:57:25 -08:00
generator mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
include/mlx/c MLX, MLX-C: version bump (#19647) 2026-11-10 13:42:11 -08:00
mlxtest mlx: System one support (#18701) 2026-10-03 13:17:21 -07:00
mlxthread mlx: System one support (#18701) 2026-10-03 13:17:21 -07:00
quant mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
.gitignore mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
act.go mlx: speed up Qwen 3.8 prompt processing (#18550) 2026-09-22 17:53:23 -07:00
act_test.go mlx: speed up Qwen 3.8 prompt processing (#18550) 2026-09-22 17:53:23 -07:00
array.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
array_test.go mlx: compact restored recurrent state after speculative rollback (#18955) 2026-10-12 09:28:22 -07:00
bf16_matmul.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
bf16_matmul_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
block_residual_mix.go mlx: fuse Inkling residual mixing on Metal (#19808) 2026-11-20 20:05:50 -08:00
block_residual_mix_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
CMakeLists.txt mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
compile.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
compile_test.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
cuda.go mlx: reject unsupported CUDA devices (#19521) 2026-11-04 09:57:25 -08:00
cuda_test.go mlx: reject unsupported CUDA devices (#19521) 2026-11-04 09:57:25 -08:00
depthwise_conv.go mlx: port the Metal custom kernels to CUDA (#18657) 2026-10-13 15:12:17 -07:00
depthwise_conv_test.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
dtype.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
dynamic.c mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
dynamic.go mlx: load bundled CUDA libraries in-process on Linux (#19433) 2026-11-03 06:45:30 -08:00
dynamic.h mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
dynamic_darwin.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
dynamic_other.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
fast.go MLX, MLX-C: version bump - prompt speed up for gemma/flint/laguna (#19291) 2026-10-25 21:25:58 -07:00
fast_gated_delta_test.go MLX: version bump (#18576) 2026-09-22 17:13:16 -07:00
gated_delta.go mlx: port the Metal custom kernels to CUDA (#18657) 2026-10-13 15:12:17 -07:00
gated_delta_test.go mlx: port the Metal custom kernels to CUDA (#18657) 2026-10-13 15:12:17 -07:00
generated.c MLX, MLX-C: version bump (#19647) 2026-11-10 13:42:11 -08:00
generated.h MLX, MLX-C: version bump (#19647) 2026-11-10 13:42:11 -08:00
gpu_kernel.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
indexed_attention.go MLX: make Qwen3.8 Flash Next memory efficient (#18078) 2026-11-06 13:39:21 -08:00
indexed_attention_test.go MLX: make Qwen3.8 Flash Next memory efficient (#18078) 2026-11-06 13:39:21 -08:00
io.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
io_test.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
kernel_device.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
kernel_device_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
kernel_test.go mlx: port the Metal custom kernels to CUDA (#18657) 2026-10-13 15:12:17 -07:00
mamba2_scan.go New model... (#19030) 2026-10-15 07:28:26 -07:00
mamba_kernels_test.go New model... (#19030) 2026-10-15 07:28:26 -07:00
memory.go mlxrunner: adapt saved history to memory pressure (#19678) 2026-11-19 15:41:42 -08:00
memory_test.go mlxrunner: adapt saved history to memory pressure (#19678) 2026-11-19 15:41:42 -08:00
mlx.go mlx: reject unsupported CUDA devices (#19521) 2026-11-04 09:57:25 -08:00
nn.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
nvfp4_activation.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
nvfp4_activation_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
nvfp4_matmul.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
nvfp4_matmul_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
nvfp4_matvec.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
nvfp4_matvec_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
ops.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
ops_extra.go mlx: speed up Qwen 3.8 prompt processing (#18550) 2026-09-22 17:53:23 -07:00
ops_extra_test.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
path.go mlx: load bundled CUDA libraries in-process on Linux (#19433) 2026-11-03 06:45:30 -08:00
path_linux_test.go mlx: load bundled CUDA libraries in-process on Linux (#19433) 2026-11-03 06:45:30 -08:00
path_test.go mlx: deduplicate CUDA runtime payloads (#17956) 2026-10-20 14:15:15 -07:00
random.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
relative_attention.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
relative_attention_prefill.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
relative_attention_prefill_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
relative_attention_probabilities_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
relative_attention_test.go mlxrunner: add Inkling 2 Mini (#19799) 2026-11-20 18:38:44 -08:00
scope.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
scope_test.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
slice.go mlx, mlxrunner: move the MLX engine out of x/ 2026-09-16 14:06:08 -07:00
stream.go Reduce model lookup and MLX decision request overhead (#18806) 2026-10-05 17:34:41 -07:00
thread_test.go MLX: version bump (#18720) 2026-10-05 17:41:51 -07:00
topk.go mlx: sort long top-k rows in chunks (#19392) 2026-10-29 18:53:49 -07:00
topk_test.go mlx: sort long top-k rows in chunks (#19392) 2026-10-29 18:53:49 -07:00
weighted_sum.go New model... (#19030) 2026-10-15 07:28:26 -07:00
weighted_sum_test.go New model... (#19030) 2026-10-15 07:28:26 -07:00