Loading activity…
@sadhikamalladi
faculty at UCSD CSE
AI
A researcher says frameworks for deriving batch-size scaling rules are less general than they claim.
AI
Its team claims activation perturbations can approach backprop, but an outside researcher questions the evidence.
AI
A user’s summary says TailSFT records each training example’s loss before supervised fine-tuning, then excludes the most-improved examples in each batch from model updates.
