3 ms·
Now we just need someone to figure out how to compress the model to get similar performance in 10B parameters. I assume some of the services that offer GPT-J A
by wongarsu 4y ago
Now we just need someone to figure out how to compress the model to get similar performance in 10B parameters.
I assume some of the services that offer GPT-J APIs will pick this up, but it doesn't look cheap or easy to get this running.