Replacing the parameter count in a Chinchilla-style scaling law with the average active parameter count during pre-training predicts final loss for both dense and sparsely pre-trained LLMs.
Data scaling laws in NMT : The effect of noise and architecture
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
-
The Journey Matters: Average Parameter Count over Pre-training Unifies Sparse and Dense Scaling Laws
Replacing the parameter count in a Chinchilla-style scaling law with the average active parameter count during pre-training predicts final loss for both dense and sparsely pre-trained LLMs.