4 ms·
There’s not enough improvement over regular LLMs to motivate optimization effort. Recall that the original transformer was well received because it was fast and
by kadushka 2y ago
There’s not enough improvement over regular LLMs to motivate optimization effort. Recall that the original transformer was well received because it was fast and scalable compared to RNNs.