denocris's picture
Update README.md
e192953
|
raw
history blame
1.19 kB
metadata
language: it
license: mit
widget:
  - text: Per fare la carbonara servono le [MASK].
  - text: La pasta più semplice è aglio, [MASK] e peperoncino.
  - text: A tavola non può mancare del buon [MASK].

ChefBERTo 👨‍🍳

chefberto-italian-cased is a BERT model obtained by MLM adaptive-tuning bert-base-italian-xxl-cased on Italian cooking recipes, approximately 50k sentences (2.6M words).

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


Perplexity

Test set: 9k sentences about food.

Model Perplexity
chefberto-italian-cased 1.84
bert-base-italian-xxl-cased 2.85

Usage

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