Compatibility check
Can I run Phi 4 on Apple M1 Ultra (64-core GPU)?
Yes — it runs.best quant:
Decode speed
63 tok/sest
Usable context
16K
Memory
128 GB
Fits in memory63 tok/sest · 16K ctx
Fits in memory54 tok/sest · 16K ctx
Fits in memory36 tok/sest · 16K ctx
Fits in memory19 tok/sest · 16K ctx
Fits in memory · 16K ctx
