metadata

language:
  - it
license: mit
tags:
  - generated_from_trainer
datasets:
  - tner/wikiann
metrics:
  - precision
  - recall
  - f1
  - accuracy
widget:
  - text: 'Ciao, sono Giacomo. Vivo a Milano e lavoro da Armani. '
    example_title: Example 1
  - text: 'Domenica andrò allo stadio con Giovanna a guardare la Fiorentina. '
    example_title: Example 2
base_model: dbmdz/bert-base-italian-cased
model-index:
  - name: bert-italian-finetuned-ner
    results:
      - task:
          type: token-classification
          name: Token Classification
        dataset:
          name: wiki_neural
          type: wiki_neural
          config: it
          split: validation
          args: it
        metrics:
          - type: precision
            value: 0.9438064759036144
            name: Precision
          - type: recall
            value: 0.954225352112676
            name: Recall
          - type: f1
            value: 0.9489873178118493
            name: F1
          - type: accuracy
            value: 0.9917883014379933
            name: Accuracy

bert-italian-finetuned-ner

This model is a fine-tuned version of dbmdz/bert-base-italian-cased on the wiki_neural dataset. It achieves the following results on the evaluation set:

Loss: 0.0361
Precision: 0.9438
Recall: 0.9542
F1: 0.9490
Accuracy: 0.9918

Model description

Token classification for italian language experiment, NER.

Example

from transformers import pipeline
ner_pipeline = pipeline("ner", model="nickprock/bert-italian-finetuned-ner", aggregation_strategy="simple")
text = "La sede storica della Olivetti è ad Ivrea"
output = ner_pipeline(text)

Intended uses & limitations

The model can be used on token classification, in particular NER. It is fine tuned on italian language.

Training and evaluation data

The dataset used is wikiann

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.0297	1.0	11050	0.0323	0.9324	0.9420	0.9372	0.9908
0.0173	2.0	22100	0.0324	0.9445	0.9514	0.9479	0.9915
0.0057	3.0	33150	0.0361	0.9438	0.9542	0.9490	0.9918

Framework versions

Transformers 4.27.3
Pytorch 1.13.0
Datasets 2.1.0
Tokenizers 0.13.2