alt Hacker NewsThe quality of custom models trained with proper reasoning datasets[0] even with small parameters (3-7B is sweet spot) is incredible now
[0]: cartesien.io or Salesforce's WebscaleRL
The quality of custom models trained with proper reasoning datasets[0] even with small parameters (3-7B is sweet spot) is incredible now
[0]: cartesien.io or Salesforce's WebscaleRL
What are you basing how good they are on? Personal experience or some benchmarks?