Mean pooling — mask-aware aggregation of hidden states.
Computes: pooled[b][d] = sum_s hidden[b][s][d] * mask[b][s] / sum_s mask[b][s]
Mean pooling — mask-aware aggregation of hidden states.
Computes: pooled[b][d] = sum_s hidden[b][s][d] * mask[b][s] / sum_s mask[b][s]