3 ms·
One wrinkle, is that it is now common to fine-tune on previously derived RL datasets, with the tested inputs and preferred sample outputs as the training data.
by jacobr1 2y ago
One wrinkle, is that it is now common to fine-tune on previously derived RL datasets, with the tested inputs and preferred sample outputs as the training data.