FinBERT Large Toxicity
This Finnish classifier was fine-tuned on translated Jigsaw comments. Its six outputs can apply together because inference uses sigmoid scores.
The trained task
The card's introduction names FinBERT Base, but its training settings name FinBERT Large. The released configuration confirms 24 layers and hidden width 1,024. The outputs cover identity attack, insult, obscene language, severe toxicity, threat and toxicity.FinBERT Large Toxicity: model card and trained artifactsFinBERT Large Toxicity: output labels and configuration
Practical assessment
Practical guidance. Set thresholds using labelled examples from the discussion setting where the model will run. Translated comments can differ from original Finnish comments. Keep each label score instead of reducing every case to the largest output.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | bert-large-finnish-cased-toxicity |
| Architecture | BERT sequence classifier; 24 layers, hidden width 1,024.FinBERT Large Toxicity: output labels and configuration |
| Output scheme | Six labels with independent probabilities.FinBERT Large Toxicity: output labels and configuration |
| License | The reviewed derivative card does not state a weight license.FinBERT Large Toxicity: model card and trained artifacts |
Related models
- FinBERT Large Cased
TurkuNLP ยท FinBERT
FinBERT Large Cased increases encoder depth and width while using the same Finnish training data as FinBERT Base.
Launch date not confirmedDownloadable