3 ms·
Not really. They quantized the activations here with their inference program which decreased compute as well as RAM usage (and required bandwidth). That's a big
by superkuh 2y ago
Not really. They quantized the activations here with their inference program which decreased compute as well as RAM usage (and required bandwidth). That's a big step.