Re:Llama3 implemented from scratch
by fire_lake on Monday May 20, 2024 @01:26PM
(#40417822)
Hard to defend because once your model is out there other companies can train on its output.
[ Parent
| on HN ]
Re:Llama3 implemented from scratch
by stephc_int13 on Monday May 20, 2024 @07:33PM
(#40421947)
Yes, but the output is only one third of the data. You also need the input and the annotations.