DeBERTa-v3

GPTKB entity

Statements (34)
Predicate Object
gptkbp:instanceOf large language model
gptkbp:author gptkb:Pengcheng_He
gptkb:Weizhu_Chen
gptkb:Xiaodong_Liu
gptkb:Jianfeng_Gao
gptkbp:availableOn gptkb:Hugging_Face_Model_Hub
gptkbp:basedOn Transformer architecture
gptkbp:bench gptkb:GLUE
gptkb:SQuAD
gptkb:SuperGLUE
gptkbp:citation 2021
gptkbp:developedBy gptkb:Microsoft_Research
gptkbp:feature disentangled attention mechanism
enhanced mask decoder
improved parameter efficiency
https://www.w3.org/2000/01/rdf-schema#label DeBERTa-v3
gptkbp:improves gptkb:BERT
gptkb:DeBERTa
gptkb:RoBERTa
gptkbp:language English
gptkbp:license gptkb:MIT_License
gptkbp:notablePublication DeBERTaV3: Improving DeBERTa using ELECTRA-Style Pre-Training with Gradient Disentangled Embedding Sharing
gptkbp:openSource true
gptkbp:parameter varies by model size
gptkbp:predecessor gptkb:DeBERTa
gptkbp:pretrainingMethod ELECTRA-style pre-training
gptkbp:releaseYear 2021
gptkbp:usedFor natural language processing
question answering
text generation
text classification
named entity recognition
gptkbp:bfsParent gptkb:DeBERTa
gptkbp:bfsLayer 6