# German NER model

**URL:** <https://support.prodi.gy/t/german-ner-model/3653>\
**Category:** Uncategorized\
**Tags:** usage, spacy\
**Created:** [November 22, 2020, 8:53pm UTC](https://support.prodi.gy/t/german-ner-model/3653 "2020-11-22T20:53:33Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![mystuff](https://avatars.discourse-cdn.com/v4/letter/m/a88e4f/32.png) [@mystuff](https://support.prodi.gy/u/mystuff)\
**Post date:** [November 22, 2020, 8:53pm UTC](https://support.prodi.gy/t/german-ner-model/3653/1 "2020-11-22T20:53:33Z")

</div>

Hi @ines,  
I have gathered 15k dataset in German language. I ran prodigy ner command with different models. I didn't find vectors model in German so used de large one. I was expecting higher accuracy using de\_core\_news\_lg than en\_vectors because of the language. May I know why it has 4% difference?

en\_vectors\_web\_lg - 92%  
de\_core\_news\_lg - 88%

Thanks,

---

<div class="post-metadata">

**Author:** ![honnibal](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/honnibal/32/35_2.png) [@honnibal](https://support.prodi.gy/u/honnibal)\
**Post date:** [November 23, 2020, 6:20am UTC](https://support.prodi.gy/t/german-ner-model/3653/2 "2020-11-23T06:20:09Z")

</div>

Hi @mystuff,

I think the problem might be that you're "resuming" the weights from `de_core_news_lg`, instead of starting from new weights and just using the vectors. Can you paste the commands you used for each model?

---

<div class="post-metadata">

**Author:** ![mystuff](https://avatars.discourse-cdn.com/v4/letter/m/a88e4f/32.png) [@mystuff](https://support.prodi.gy/u/mystuff)\
**Post date:** [November 23, 2020, 9:23am UTC](https://support.prodi.gy/t/german-ner-model/3653/3 "2020-11-23T09:23:12Z")

</div>

Hi @honnibal,  
Thanks for the reply

python -m prodigy train ner de\_20000 de\_core\_news\_lg --output de\_20000\_core\_lg  
python -m prodigy train ner de\_20000 en\_vectors\_web\_lg --output de\_20000\_en\_vectors

---

<div class="post-metadata">

**Author:** ![honnibal](https://sea2.discourse-cdn.com/flex020/user_avatar/support.prodi.gy/honnibal/32/35_2.png) [@honnibal](https://support.prodi.gy/u/honnibal)\
**Post date:** [November 26, 2020, 1:24am UTC](https://support.prodi.gy/t/german-ner-model/3653/4 "2020-11-26T01:24:07Z")

</div>

Yes I do think the issue is the weight resuming. You can create a model that has the vectors but dumps the trained pipeline components like this:

```python
import spacy
nlp = spacy.load("de_core_news_lg")
with nlp.disable_pipes(*nlp.pipe_names):
    nlp.to_disk("./de_vectors_news_lg")

```

Alternatively, let's say you want to make a new model, with your own vectors. You can do that with the command `spacy init-model`, as described here: [https://spacy.io/usage/vectors-similarity#converting](https://spacy.io/usage/vectors-similarity#converting) . You can convert vectors from tools like FastText, so you could use the trained models from fasttext.cc