Adding a CTC-style <SLACK> token during MDLM supervised fine-tuning relaxes strict positional supervision and improves open-ended generation benchmarks over a matched cross-entropy baseline.
Title resolution pending
1 Pith paper cite this work, alongside 64 external citations. Polarity classification is still indexing.
1
Pith paper citing it
64
external citations · OpenAlex
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Reconsidering Positional Supervision in Masked Diffusion Language Model Training
Adding a CTC-style <SLACK> token during MDLM supervised fine-tuning relaxes strict positional supervision and improves open-ended generation benchmarks over a matched cross-entropy baseline.