denocris's picture
Update README.md
3e2bacd
|
raw
history blame
1.21 kB
metadata
language: it
license: mit
widget:
  - text: Con del pesce bisogna bere un bicchiere di vino [MASK].
  - text: Con la carne c'è bisogno del vino [MASK].
  - text: A tavola non può mancare del buon [MASK].

WineBERTo 🍷🥂

wineberto-italian-cased is a BERT model obtained by MLM adaptive-tuning bert-base-italian-xxl-cased on Italian drink recipes and wine descriptions, approximately 77k sentences (3.3M words).

Author: Cristiano De Nobili (@denocris on Twitter, LinkedIn) for VINHOOD.


Perplexity

Test set: 14k sentences about wine.

Model Perplexity
wineberto-italian-cased 2.28
bert-base-italian-xxl-cased 4.60

Usage

from transformers import AutoModel, AutoTokenizer
model_name = "vinhood/wineberto-italian-cased"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name)