FinBERT Large Cased
FinBERT Large Cased is the larger Finnish BERT release. Its card says it uses exactly the same training data as FinBERT Base Cased.
A larger encoder, not a longer input window
The released configuration doubles the base model's layer count to 24 and raises the hidden width to 1,024. Its position limit remains 512. The model loads with a masked-language-model head.FinBERT Large Cased: configuration
Choose capacity for the task
Practical guidance. A larger encoder may change fine-tuning memory needs without allowing longer documents. Compare the base and large releases on the same labelled examples, and keep long-document splitting separate from model selection.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | bert-large-finnish-cased-v1 |
| Context window | 512 tokens Position limit in the released encoder configuration. FinBERT Large Cased: configuration |
| License | Apache 2.0 in this release's metadata.FinBERT Large Cased: model card |
| Architecture | 24 encoder layers, hidden width 1,024 and 16 attention heads.FinBERT Large Cased: configuration |
| Vocabulary | 50,105 entries; case-sensitive Finnish release.FinBERT Large Cased: configuration |
Related models
- FinBERT Base Cased
TurkuNLP · FinBERT
FinBERT Base Cased is a Finnish masked-language encoder with a dedicated WordPiece vocabulary.
Launch date Downloadable - Megatron FinBERT 1.3B
TurkuNLP · FinBERT
Megatron FinBERT is a 1.3-billion-parameter Finnish encoder with a wider Megatron-BERT architecture.
Launch date not confirmedDownloadable