TensorFlow Lite Swift XCFramework Distribution

A Swift Package Manager (SwiftPM) distribution of TensorFlow Lite (Swift), packaged from the official Google prebuilt binary.

Why This Repository?

TensorFlow Lite (Swift) has long been available only via CocoaPods, with no official SwiftPM distribution (google-ai-edge/LiteRT#125). This repository re-packages the same binary that CocoaPods downloads, together with the official Swift wrapper sources, as a SwiftPM package so it can be consumed through SwiftPM.

The approach mirrors mitene/realm-swift-xcframework-distribution.

Installation

Swift Package Manager

Add the package as a dependency:

dependencies: [
    .package(url: "https://github.com/mitene/tensorflow-lite-swift-xcframework-distribution", from: "2.17.0"),
],

And add the product to your target:

.product(name: "TensorFlowLite", package: "tensorflow-lite-swift-xcframework-distribution")

import TensorFlowLite

let interpreter = try Interpreter(modelPath: modelPath)

binaryTarget(url:) downloads the zip once at package-resolution time and caches it locally afterwards (not on every build).

What's Included

.
├─ Package.swift                     # TensorFlowLite library (Swift wrapper + TensorFlowLiteC binaryTarget)
├─ Sources/TensorFlowLite/           # Swift wrapper + PrivacyInfo.xcprivacy
├─ LICENSE
└─ README.md

Release v2.17.0
└─ TensorFlowLiteC.xcframework.zip   # Google prebuilt (Core only); Release asset, not tracked in git

Only the Core framework is included; the CoreML and Metal delegates are excluded.

Provenance

Because a re-packaged binary loses its trail back to the original, the origin of both components is recorded here.

TensorFlowLiteC (binary)

• Origin: the Google official prebuilt fetched by CocoaPods TensorFlowLiteC 2.17.0 (a dl.google.com tar.gz, Apache-2.0).
• Source URL:

  https://dl.google.com/tflite-release/ios/prod/tensorflow/lite/release/ios/release/32/20240729-115310/TensorFlowLiteC/2.17.0/0c10b3543e01f547/TensorFlowLiteC-2.17.0.tar.gz
  

• Modifications: only a minimal Info.plist is added to each framework slice (ios-arm64, ios-arm64_x86_64-simulator) to avoid an embedding validation error. The mach-o binary, headers, module map, and PrivacyInfo.xcprivacy are unchanged from upstream.
• Binary SHA-256 (verified identical to the Google prebuilt):

  slice	SHA-256
  ios-arm64	0af9943357e8b94e16b28f807d9acf44f596954191e34b4f5dd46d0c49b29561
  ios-arm64_x86_64-simulator	26a2e0179f2899e2652a76890b98b56647462add7b4d70c9bb8124dc98e66def

• Release zip SwiftPM checksum: 8da2a4112bd0c65ab08fa037ffee61d96f265290e885413e61cd41d190b37587

TensorFlowLite (Swift sources)

• Origin: tensorflow/tensorflow @ commit ad6d8cc177d0c868982e39e0823d0efbfb95f04c, tensorflow/lite/swift/Sources/*.swift.
• Excluded: CoreMLDelegate.swift and MetalDelegate.swift.
• Module name: TensorFlowLite.

Regenerating the Release Asset

The TensorFlowLiteC.xcframework.zip on the Release can be reproduced as follows:

# 1. Fetch the Google official tar.gz
curl -fL -o TensorFlowLiteC-2.17.0.tar.gz \
  "https://dl.google.com/tflite-release/ios/prod/tensorflow/lite/release/ios/release/32/20240729-115310/TensorFlowLiteC/2.17.0/0c10b3543e01f547/TensorFlowLiteC-2.17.0.tar.gz"
tar -xzf TensorFlowLiteC-2.17.0.tar.gz

# 2. Keep only the Core framework (drop CoreML / Metal)
cd TensorFlowLiteC-2.17.0/Frameworks

# 3. Add a minimal Info.plist to each slice
#    (CFBundleIdentifier=org.tensorflow.TensorFlowLiteC, CFBundleExecutable=TensorFlowLiteC,
#     CFBundlePackageType=FMWK, MinimumOSVersion=12.0, ...)

# 4. Zip and compute the checksum
zip -r -X TensorFlowLiteC.xcframework.zip TensorFlowLiteC.xcframework
swift package compute-checksum TensorFlowLiteC.xcframework.zip

License

• TensorFlow Lite (both the binary and the Swift sources): Apache License 2.0. See LICENSE. Copyright belongs to The TensorFlow Authors and contributors.