Token superposition in an initial training phase followed by recovery allows large language models to reach target loss with substantially less total compute.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Efficient Pre-Training with Token Superposition
Token superposition in an initial training phase followed by recovery allows large language models to reach target loss with substantially less total compute.