BWhat changed
Separated token content and position representations in attention and used an enhanced mask decoder, reporting improvements over BERT/RoBERTa baselines.
Scientific scope
Separated token content and position representations in attention and used an enhanced mask decoder, reporting improvements over BERT/RoBERTa baselines. Scope limitation: The 2020 preprint was published at ICLR 2021; benchmark gains depend on the paper's training and evaluation choices.
Evidence
The retained primary-source abstract, bibliographic record and stated review scope support this provisional assessment. No independent experimental replication by WRS is claimed. The 2020 preprint was published at ICLR 2021; benchmark gains depend on the paper's training and evaluation choices.
Contributors
- The named authors listed in the linked research record · SCIENCE: Complete source byline retained in the research record; human research direction, method development and evaluation. Individual labor shares are not measured.
- He and colleagues: DeBERTa disentangled attention — original paper
- He and colleagues: DeBERTa disentangled attention — official company source
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction