forward method
Implementation
Tensor forward(Tensor x, List<Tensor> tracker) {
// 1. Compute each head on GPU
final List<Tensor> headOutputs = heads
.map((h) => h.forward(x, tracker))
.toList();
// 2. Concatenate results on GPU
final concatenated = Tensor.concat(headOutputs);
tracker.add(concatenated);
// 3. Final Linear Projection
return proj.forward(concatenated, tracker);
}