Combining a frozen MLM and AR model with a trained linear layer enables autoregressive models to perform masked token infilling with KV-cached inference.
One billion word benchmark for measuring progress in statistical language modeling, 2014
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
-
Enabling Autoregressive Models to Fill In Masked Tokens
Combining a frozen MLM and AR model with a trained linear layer enables autoregressive models to perform masked token infilling with KV-cached inference.