tensorflow #3246

TheCodeCache · 2024-05-08T21:55:40Z

TheCodeCache
May 8, 2024

Answered by ndonkoHenri

Unfortunately, full TensorFlow currently can't be packaged for a Flet mobile app:

No source distribution on PyPi: Every TensorFlow release ships only pre-built platform wheels (Linux/macOS/Windows) — there's no sdist, so there's nothing for Flet's mobile build to cross-compile from.
Bazel-only build: TensorFlow is built exclusively with Bazel, not setuptools/CMake, so the iOS/Android cross-compile toolchain Flet uses can't drive its build.
Size: The compiled native runtime (libtensorflow + XLA + MLIR + Eigen + abseil) is ~220–570 MB per platform — far beyond what's reasonable to embed in an app?

Possible alternative: run inference on a backend service and call it from the app.

ndonkoHenri · 2026-07-02T15:31:07Z

Unfortunately, full TensorFlow currently can't be packaged for a Flet mobile app:

No source distribution on PyPi: Every TensorFlow release ships only pre-built platform wheels (Linux/macOS/Windows) — there's no sdist, so there's nothing for Flet's mobile build to cross-compile from.
Bazel-only build: TensorFlow is built exclusively with Bazel, not setuptools/CMake, so the iOS/Android cross-compile toolchain Flet uses can't drive its build.
Size: The compiled native runtime (libtensorflow + XLA + MLIR + Eigen + abseil) is ~220–570 MB per platform — far beyond what's reasonable to embed in an app?

Possible alternative: run inference on a backend service and call it from the app.

0 replies