Patrick Devine 710292ff4f mlx: tighten up gemma4 moe loading code (#16964) ...

This change allows .experts.gate_proj / .up_proj / .down_proj tensor names to each
be used for both quantized (i.e. nvfp4 and mxfp8) and non-quantized (bf16) models.
Previous to this only non-quantized models used that tensor naming scheme.

2026-06-29 21:15:08 -07:00

..

agent

x/cmd: enable web search and web fetch with flag (#13690)

2026-01-12 13:59:40 -08:00

cmd

Reapply "don't require pulling stubs for cloud models" again (#14608)

2026-03-06 14:27:47 -08:00

create

models: add cohere2_moe (Command A / North) to the MLX engine (#16670)

2026-06-16 23:15:21 -07:00

imagegen

mlx: update and fix CUDA JIT packaging (#16871)

2026-06-24 10:36:02 -07:00

internal/mlxthread

mlxthread: preserve the original stack when worker work panics

2026-06-09 00:39:19 -07:00

mlxrunner

mlx: bump dependency (#16935)

2026-06-29 09:39:11 -07:00

models

mlx: tighten up gemma4 moe loading code (#16964)

2026-06-29 21:15:08 -07:00

safetensors

mlx: Support NVIDIA TensorRT Model Optimizer import (#15566)

2026-04-27 18:28:10 -07:00

server

mlx: fix reported information in ollama show (#16289)

2026-05-24 14:08:06 -07:00

tokenizer

runner: Remove CGO engines, use llama-server exclusively for GGML models (#16031)

2026-05-29 13:35:47 -07:00

tools

add ability to disable cloud (#14221)

2026-02-12 15:47:00 -08:00

transfer

mlx: refined model push behavior (#15431)

2026-05-08 14:25:30 -07:00