Could you make the TASO support softmax operator given the popularity of it? We want to run TASO with the transformer architecture or the entire ResNet network, but lack of softmax support becomes a bottleneck and I believe we are not the only one who wants to run the TASO with such popular models.
Could you make the TASO support softmax operator given the popularity of it? We want to run TASO with the transformer architecture or the entire ResNet network, but lack of softmax support becomes a bottleneck and I believe we are not the only one who wants to run the TASO with such popular models.