A lower learning rate on correct examples after training on errors lets Gemma 2 set a new single-model SOTA on BEA-test, aided by adding unedited pairs.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Adapting LLMs for Minimal-edit Grammatical Error Correction
A lower learning rate on correct examples after training on errors lets Gemma 2 set a new single-model SOTA on BEA-test, aided by adding unedited pairs.