spaCy/website/docs/usage/_benchmarks-models.md

48 lines
2.8 KiB
Markdown
Raw Normal View History

2020-09-12 18:05:10 +03:00
import { Help } from 'components/typography'; import Link from 'components/link'
2020-10-14 15:58:45 +03:00
<!-- TODO: update speed and v2 NER numbers -->
2020-09-12 18:05:10 +03:00
<figure>
| Pipeline | Parser | Tagger | NER | WPS<br />CPU <Help>words per second on CPU, higher is better</Help> | WPS<br/>GPU <Help>words per second on GPU, higher is better</Help> |
2020-09-23 23:02:31 +03:00
| ---------------------------------------------------------- | -----: | -----: | ---: | ------------------------------------------------------------------: | -----------------------------------------------------------------: |
2020-10-15 13:35:30 +03:00
| [`en_core_web_trf`](/models/en#en_core_web_trf) (spaCy v3) | 95.5 | 98.3 | 89.4 | 1k | 8k |
2020-10-15 12:16:06 +03:00
| [`en_core_web_lg`](/models/en#en_core_web_lg) (spaCy v3) | 92.2 | 97.4 | 85.4 | 7k | |
| `en_core_web_lg` (spaCy v2) | 91.9 | 97.2 | 85.7 | 10k | |
2020-09-12 18:05:10 +03:00
<figcaption class="caption">
2020-09-24 11:13:41 +03:00
**Full pipeline accuracy and speed** on the
2020-10-15 12:16:06 +03:00
[OntoNotes 5.0](https://catalog.ldc.upenn.edu/LDC2013T19) corpus (reported on
the development set).
2020-09-12 18:05:10 +03:00
</figcaption>
</figure>
<figure>
| Named Entity Recognition System | OntoNotes | CoNLL '03 |
2020-09-24 13:37:21 +03:00
| ------------------------------------------------------------------------------ | --------: | --------: |
2020-10-14 15:58:45 +03:00
| spaCy RoBERTa (2020) | 89.7 | 91.6 |
2020-10-15 09:58:30 +03:00
| spaCy CNN (2020) | 84.5 | 87.4 |
2020-09-24 13:37:21 +03:00
| [Stanza](https://stanfordnlp.github.io/stanza/) (StanfordNLP)<sup>1</sup> | 88.8 | 92.1 |
| <Link to="https://github.com/flairNLP/flair" hideIcon>Flair</Link><sup>2</sup> | 89.7 | 93.1 |
| BERT Base<sup>3</sup> | - | 92.4 |
2020-09-12 18:05:10 +03:00
<figcaption class="caption">
2020-09-23 23:02:31 +03:00
**Named entity recognition accuracy** on the
[OntoNotes 5.0](https://catalog.ldc.upenn.edu/LDC2013T19) and
[CoNLL-2003](https://www.aclweb.org/anthology/W03-0419.pdf) corpora. See
[NLP-progress](http://nlpprogress.com/english/named_entity_recognition.html) for
2020-10-15 09:58:30 +03:00
more results. Project template:
[`benchmarks/ner_conll03`](%%GITHUB_PROJECTS/benchmarks/ner_conll03). **1. **
[Qi et al. (2020)](https://arxiv.org/pdf/2003.07082.pdf). **2. **
[Akbik et al. (2018)](https://www.aclweb.org/anthology/C18-1139/). **3. **
[Devlin et al. (2018)](https://arxiv.org/abs/1810.04805).
2020-09-12 18:05:10 +03:00
</figcaption>
</figure>