Running P parallel streams of a language model with learned prefixes gives performance comparable to multiplying parameters by about k log P + 1, with k around 0.33 to 0.39.
Speeding up Deep Model Training by Sharing Weights and Then Unsharing
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We propose a simple and efficient approach for training the BERT model. Our approach exploits the special structure of BERT that contains a stack of repeated modules (i.e., transformer encoders). Our proposed approach first trains BERT with the weights shared across all the repeated modules till some point. This is for learning the commonly shared component of weights across all repeated layers. We then stop weight sharing and continue training until convergence. We present theoretic insights for training by sharing weights then unsharing with analysis for simplified models. Empirical experiments on the BERT model show that our method yields better performance of trained models, and significantly reduces the number of training iterations.
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Parallel Scaling Law for Language Models
Running P parallel streams of a language model with learned prefixes gives performance comparable to multiplying parameters by about k log P + 1, with k around 0.33 to 0.39.