NER4Archives pipeline optimized for CPU and specialized on French National Archives findings aids (XML-EAD) - Corpus V2. Components: tok2vec, ner. Base default CNN architecture.
| Feature | Description |
|---|---|
| Name | fr_ner4archives_default_test |
| Version | 0.0.0 |
| spaCy | >=3.3.1,<3.4.0 |
| Default Pipeline | tok2vec, ner |
| Components | tok2vec, ner |
| Vectors | 0 keys, 0 unique vectors (0 dimensions) |
| Sources | French corpus for the NER task composed of finding aids in XML-EAD from the National Archives of France (v. 2.0) - Check corpus version on GitHub |
| License | CC-BY-4.0 license |
| Author | Archives nationales / Inria-Almanach |
Label Scheme
View label scheme (5 labels for 1 components)
| Component | Labels |
|---|---|
ner |
EVENT, LOCATION, ORGANISATION, PERSON, TITLE |
Accuracy
| Type | Score |
|---|---|
ENTS_F |
76.95 |
ENTS_P |
80.00 |
ENTS_R |
74.13 |
TOK2VEC_LOSS |
76044.50 |
NER_LOSS |
75529.77 |
- Downloads last month
- 4
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
Evaluation results
- NER Precisionself-reported0.800
- NER Recallself-reported0.741
- NER F Scoreself-reported0.770