ModelCheckpoint class

Dart-side weight persistence for on-device training models.

Saves and restores model weights to disk using the get_weights and set_weights signatures that training models expose. This is exported on native platforms only and does not require the Flex delegate or any native library beyond the base TFLite runtime.

Binary checkpoint format (.flwt)

[4B magic "FLWT"] [1B version] [4B tensor count N]
Per tensor:
  [4B name length] [name UTF-8] [4B TfLiteType] [4B rank]
  [rank*4B shape dims] [4B data byte count] [data bytes]

Example

final interpreter = Interpreter.fromFile(modelFile);

// Train...
final trainRunner = interpreter.getSignatureRunner('train');
for (int i = 0; i < 100; i++) {
  trainRunner.run({'x': [[1.0]], 'y': [[2.0]]}, {'loss': Float32List(1)});
}
trainRunner.close();

// Save weights to disk
await ModelCheckpoint.save(interpreter, File('model.flwt'));

// Later, restore into a fresh interpreter
final fresh = Interpreter.fromFile(modelFile);
await ModelCheckpoint.restore(fresh, File('model.flwt'));

Properties

hashCode → int

The hash code for this object.

no setter, inherited

runtimeType → Type

A representation of the runtime type of the object.

no setter, inherited

Methods

noSuchMethod(Invocation invocation) → dynamic

Invoked when a nonexistent method or property is accessed.

inherited

toString() → String

A string representation of this object.

inherited

Operators

operator ==(Object other) → bool

The equality operator.

inherited

Static Methods

inspect(File file) → Future<Map<String, List<int>>>

Returns a map of tensor name to shape for every tensor in a checkpoint file. Reads and parses the entire file, then discards the tensor data, retaining only names and shapes. Useful for debugging and validation.

restore(Interpreter interpreter, File file) → Future<void>

Restores model weights from file into interpreter.

save(Interpreter interpreter, File file) → Future<void>

Saves the current model weights to file.