3 ms·
Without speaking to the capabilities of TPUs, note that most ML models today are mostly convolutions and matrix multiplies.
by jlebar 9y ago
Without speaking to the capabilities of TPUs, note that most ML models today are mostly convolutions and matrix multiplies.