forward method
Forward pass (no sampling): returns logits (T, vocab)
Implementation
Tensor2D forward(List<int> tokens) {
final T = tokens.length;
if (T > cfg.maxSeqLen) {
throw ArgumentError('sequence length $T exceeds max ${cfg.maxSeqLen}');
}
var x = tokEmb.lookup(tokens);
final p = posEmb.lookup(List<int>.generate(T, (i) => i));
x = Ops.add(x, p);
for (var l = 0; l < blocks.length; l++) {
x = blocks[l].forward(x, cache: null);
}
x = lnF.forward(x);
return head.forward(x);
}