A food-bank intake desk has a few hundred labeled comments. One volunteer wants to train a transformer from scratch so it will know pantries; another wants to start from a general pretrained checkpoint and measure the difference on the same split. What is the fair experiment?
Select an answer to reveal the explanation.
Short Explanation
A few hundred pantry comments, two starts: random weights versus a general pretrained checkpoint. Treat initialization as the factor. Same split, same scorer. Training from scratch “so it knows pantries” is the hypothesis to measure, not a reason to skip the pretrained arm.
Full Explanation
When labels are scarce, the transfer question is whether a pretrained start beats a random one. That is a controlled factor: same split, same scorer, only the initialization changes. Training from scratch “so it will know pantries” is a hypothesis to measure, not a reason to skip the pretrained arm. Kernels and leaked test peeks are not that factor.