Dont work
#2
by juaps - opened
You cant load it with LMStudio or oMLX.
the same for me - download is stuck at 32.4M:
hf cache ls
ID SIZE LAST_ACCESSED LAST_MODIFIED REFS
--------------------------------------- ----- -------------- -------------- ----
model/mlx-community/gemma-4-12B-it-4bit 32.4M 16 minutes ago 16 minutes ago main
model/mlx-community/gemma-4-12B-it-8bit 32.4M 13 minutes ago 13 minutes ago main
try other models
Not a broken model, probably just a mlx-vlm version thing. This build is the new gemma4_unified architecture, which mlx-vlm only added in 0.6.0. Older runtimes (e.g. oMLX 0.4.1 bundles mlx-vlm 0.5.0) fail with Model type gemma4_unified not supported. Upgrading mlx-vlm to ≥ 0.6.0 fixes it.
Can confirm - loading & running fine here afterwards.
May I ask how to upgrade the package mlx-vlm in omlx? (or even a hint would be appreciated ..).
Ok, found that this issue is already addressed in https://github.com/jundot/omlx/issues/1645.