3 ms·
We are already doing this. In fact for the next generation of frontier models the primary electricity cost is running inference to generate training data for th
by MeImCounting 2y ago
We are already doing this. In fact for the next generation of frontier models the primary electricity cost is running inference to generate training data for these models. As Llama 3 has shown us scale of training data is more important than size of model.