3 ms·
deep learning uses low precision floats sometimes as few as 8 bits are needed
by perfmode 8y ago
deep learning uses low precision floats
sometimes as few as 8 bits are needed
- jadedhacker 8y agoI think gen 1 or gen 2 of the TPU explicitly supported short ints.