3 ms·
It keeps context of 8k tokens. In pre-training you're using much more examples and network is tuned around them.
by mirekrusin 4y ago
It keeps context of 8k tokens.
In pre-training you're using much more examples and network is tuned around them.