loadModuleBinary function

Future<bool> loadModuleBinary(Module model, String filePath)

Reads a binary file and pushes the weights into GPU VRAM

Implementation

Future<bool> loadModuleBinary(Module model, String filePath) async {
  final file = File(filePath);
  if (!await file.exists()) return false;

  final Uint8List allBytes = await file.readAsBytes();
  final Float32List allFloats = allBytes.buffer.asFloat32List();
  final List<Tensor> params = model.parameters();

  // Safety Check
  final int totalExpected = params.fold(0, (sum, p) => sum + p.length);
  if (allFloats.length != totalExpected) {
    print(
      '⚠️ Mismatch! Model needs $totalExpected floats, file has ${allFloats.length}',
    );
    return false;
  }

  print('🚀 Injecting weights into GPU VRAM...');
  int offset = 0;
  for (var p in params) {
    final int len = p.length;
    // p.data (setter) triggers engine.setTensorData -> cudaMemcpyHostToDevice
    p.data = allFloats.sublist(offset, offset + len).toList();
    offset += len;
  }
  return true;
}