TinyLoRA – Learning to Reason in 13 Parameters
by sorenjan on 3/27/2026, 12:11:12 PM
https://arxiv.org/abs/2602.04118
Comments
by: a-t-c-g
The quality of custom models trained with proper reasoning datasets[0] even with small parameters (3-7B is sweet spot) is incredible now<p>[0]: cartesien.io or Salesforce's WebscaleRL
4/1/2026, 2:01:15 AM
by: measurablefunc
With four parameters I can fit an elephant, and with five I can make him wiggle his trunk so there is still room for improvement.
4/1/2026, 1:08:22 AM
by: ValveFan6969
[dead]
4/1/2026, 1:56:39 AM