Training Loss reference & dataset size
Author: VMA-001Created Aug 28, 2026Updated Aug 28, 2026
I would like to know the general training loss range/ reference while fine tuning N1.7-3B as the base model for normal pick & precision place task. I am trying to train a 27 DOF humanoid in sim for a precision placing task, its failing horribly - even for the normal picking task . I would like to know the general or average dataset size/ no of demonstration required??
Source: NVIDIA/Isaac-GR00T