Compatibility check
Can I run Phi 4 Multimodal Instruct on Apple M5 Pro (20-core GPU)?
Yes — it runs.best quant:
Decode speed
63 tok/sest
Usable context
131K
Memory
64 GB
Fits in memory63 tok/sest · 131K ctx
Fits in memory54 tok/sest · 131K ctx
Fits in memory36 tok/sest · 131K ctx