Early-exit decoders let TabPFN stop at earlier transformer layers when confidence is high, giving up to 2.2x faster tabular classification with small accuracy losses.
Attention is all you need
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Early Stopping Tabular In-Context Learning
Early-exit decoders let TabPFN stop at earlier transformer layers when confidence is high, giving up to 2.2x faster tabular classification with small accuracy losses.