179 questions
No questions match those filters.
← All questions
Training & Fine-tuningMedium· RLHF· Troubleshooting· Reinforcement Learning
Your RLVR training curve is flat and reward isn’t impro...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansA flat training curve often indicates that the problem set is not well-calibrated. Before changing the optimizer or adding a PRM, you should re-filter the problem set to ensure a better distribution of difficulty. Additionally, verify that your verifier is not silently accepting incorrect formats, which could mask underlying issues.