bert-base-cased
Fill-Mask
•
Updated
•
4.63M
•
141
BERT release
Regroups the original BERT models released by the Google team. Except for the models marked otherwise, the checkpoints support English.
Note Base BERT model, smaller variant. Trained on the "cased" dataset, meaning that it wasn't lowercase and all accents were kept. 12-layer, 768-hidden, 12-heads , 110M parameters
bert-base-uncased
Fill-Mask • Updated • 38.8M • 1.1kNote Base BERT model, smaller variant. Trained on the "uncased" dataset, meaning that it was lowercase and all accents were removed. 12-layer, 768-hidden, 12-heads , 110M parameters
bert-large-cased
Fill-Mask • Updated • 116k • 12Note Large BERT model, larger variant. Trained on the "cased" dataset, meaning that it wasn't lowercase and all accents were kept. 24-layer, 1024-hidden, 16-heads, 340M parameters
bert-large-uncased
Fill-Mask • Updated • 547k • 49Note Large BERT model, larger variant. Trained on the "uncased" dataset, meaning that it was lowercase and all accents were removed. 24-layer, 1024-hidden, 16-heads, 340M parameters
bert-base-multilingual-cased
Fill-Mask • Updated • 2.45M • 229Note Base BERT model, smaller variant. The list of supported languages is available here: https://github.com/google-research/bert/blob/master/multilingual.md#list-of-languages 104 languages, 12-layer, 768-hidden, 12-heads, 110M parameters
bert-base-chinese
Fill-Mask • Updated • 1.09M • 554Note Base BERT model, smaller variant. Chinese Simplified and Traditional, 12-layer, 768-hidden, 12-heads, 110M parameters
bert-large-cased-whole-word-masking
Fill-Mask • Updated • 914 • 4Note Large BERT model, larger variant. Trained on the "cased" dataset, meaning that it wasn't lowercase and all accents were kept. Whole word masking indicates a different preprocessing where entire words are masked rather than subwords. The BERT team reports better metrics with the wwm models. 24-layer, 1024-hidden, 16-heads, 340M parameters
bert-large-uncased-whole-word-masking
Fill-Mask • Updated • 99.8k • 10Note Large BERT model, larger variant. Trained on the "uncased" dataset, meaning that it was lowercase and all accents were removed. Whole word masking indicates a different preprocessing where entire words are masked rather than subwords. The BERT team reports better metrics with the wwm models. 24-layer, 1024-hidden, 16-heads, 340M parameters