3 ms·
Stack more GPT-3s! Have GPT-ception via stacks of multi-headed GPT blocks. I'm sure softmax attention can be modeled as a few-shot text generation problem.
by sooheon 6y ago
Stack more GPT-3s! Have GPT-ception via stacks of multi-headed GPT blocks. I'm sure softmax attention can be modeled as a few-shot text generation problem.