2 ms·
When can we try using it?? :)
by dgreensp 5y ago
When can we try using it?? :)
- gjstein 5y agoWould love an answer on this too. It would be even better not just to try using this, but also be able to run it locally, something that has been impossible for GPT-3.
- arkano 5y agoDoes it look like it would be possible to run locally?
- whimsicalism 5y agoThis is not something that will be possible to run locally. If you had 1 bit per parameter (not realistic), it would still take ~100 GB of RAM just to load into memory.
- The_rationalist 5y agoYou could technically dynamically offload the RAM overload to disk but this would probably be too slow?
- whimsicalism 5y agoI mean, theoretically if you can get the model weights onto disk then you should be able to do the computation - but it might takes days or months on commodity hardware. It would also require creating a system that can do this and I doubt there is much demand.
- ausbah 5y agoi wonder if pruning and other methods that reduce size drastically while not compromising on performance would be possible