phi-4-mini-instruct

A Core AI bundle of microsoft/Phi-4-mini-instruct for the Aether SDK (iOS and macOS 27+).

• Source: microsoft/Phi-4-mini-instruct at revision cfbefacb99257ffa30c83adab238a50856ac3083, licence MIT. The licence is included as LICENSE.
• Changes from the source: converted from PyTorch to Core AI (.aimodel) by Aether forge (recipe phi-4-mini-instruct@2). Weights are int8-linear-perblock32 (8-bit weights). The tokenizer files are the source's own.

Variants

Variant	Platform	Arch	Compute	Compiled	Assets	Download
macos-any-gpu	macos	any	gpu	no (specialized on first load)	phi_4_mini_instruct.aimodel 4.08 GB	4.1 GB
ios-h18p-gpu	ios	h18p	gpu	yes	phi_4_mini_instruct.h18p.aimodelc 4.08 GB	4.1 GB

ios-h18p-gpu needs the com.apple.developer.kernel.increased-memory-limit entitlement.

Verification

Every row is a record in verification/ about exactly these bytes (matched by bundle digest). Reference rows are strict T2 passes of the unquantized export on the same fixture, in verification/reference/.

Variant	Tier	Result	Detail	Device	OS build	Compute	Record
ios-h18p-gpu	T0	pass		iPhone18,2	24A446	target	799a7472
ios-h18p-gpu	T2	pass	20/20 strict; profile quantized-8bit; fixture e6d8aa51e5bbd5c4	iPhone18,2	24A446	target	56352841
macos-any-gpu	T0	pass		Mac17,6	26A434	target	f8ec071b
macos-any-gpu	T1	pass		Mac17,6	26A434	target	06f8b021
macos-any-gpu	T2	pass	20/20 strict; 20/20 strict; profile quantized-8bit; fixture e6d8aa51e5bbd5c4	Mac17,6	26A434	target	d86a849c
unquantized reference (not published)	T2	pass	20/20 strict; 20/20 strict; profile strict; fixture e6d8aa51e5bbd5c4	Mac17,6	26A434	target	998a7aa8

The quantized profile also requires: T2 strict on the unquantized reference export (met by the reference row).

Use

aether run phi-4-mini-instruct --prompt "Hello"

import Aether

let aether = try Aether()
let chat = try await aether.chat("phi-4-mini-instruct")
let reply = try await chat.respond(to: "Hello")
print(reply.text)