Gemma 4 26B-A4B inference in ~2 GB of RAM on any M-series MacBook
Swift apple-silicongemmagemma4gemma4-26b-a4bgpgpullmllm-inferencelocal-aimacosmetalon-device-aion-device-llmswift
import Foundation
public protocol AppVisionPackInstallerClient: Sendable {
var descriptor: AppModelInstallDescriptor { get }
func checkInstallRequirement(textModelDirectory: URL) throws -> AppModelInstallRequirement
func install(textModelDirectory: URL) -> AsyncThrowingStream<AppModelInstallEvent, Error>
func preparedInstallIsValid(textModelDirectory: URL) -> Bool
func activatePreparedInstall(
textModelDirectory: URL,
onVerifyProgress: (@Sendable (Double) -> Void)?
) async throws -> URL
func discardPartialInstall(textModelDirectory: URL) async throws
func removeInstalled(textModelDirectory: URL) async throws
func cancel()
}