Finetuning small models shifts the point where capability emerges, and extrapolating this shift to the low-data limit predicts few-shot emergence up to about 4x the compute in advance.
Emergent abilities and grokking: Fundamental, mirage, or both?, 2023
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Predicting Emergent Capabilities by Finetuning
Finetuning small models shifts the point where capability emerges, and extrapolating this shift to the low-data limit predicts few-shot emergence up to about 4x the compute in advance.