3 ms·
That may be partially because it's not possible to do pre-training on such a large model.
by mirekrusin 4y ago
That may be partially because it's not possible to do pre-training on such a large model.
- fareesh 4y agoOn the web version is it doing that under the hood? I.e. if I have a 10+ follow up conversation does it start from the beginning each time?
- mirekrusin 4y agoIt keeps context of 8k tokens. In pre-training you're using much more examples and network is tuned around them.