German Bert Embeddings (from dbmdz)

Description

Pretrained Bert Embeddings model, uploaded to Hugging Face, adapted and imported into Spark NLP. bert-base-german-uncased is a German model orginally trained by dbmdz.

Download

How to use

documentAssembler = DocumentAssembler() \
    .setInputCol("text") \
    .setOutputCol("document")

tokenizer = Tokenizer() \
    .setInputCols("document") \
    .setOutputCol("token")
  
embeddings = BertEmbeddings.pretrained("bert_embeddings_bert_base_german_uncased","de") \
    .setInputCols(["document", "token"]) \
    .setOutputCol("embeddings")
    
pipeline = Pipeline(stages=[documentAssembler, tokenizer, embeddings])

data = spark.createDataFrame([["Ich liebe Funken NLP"]]).toDF("text")

result = pipeline.fit(data).transform(data)
val documentAssembler = new DocumentAssembler() 
      .setInputCol("text") 
      .setOutputCol("document")
 
val tokenizer = new Tokenizer() 
    .setInputCols(Array("document"))
    .setOutputCol("token")

val embeddings = BertEmbeddings.pretrained("bert_embeddings_bert_base_german_uncased","de") 
    .setInputCols(Array("document", "token")) 
    .setOutputCol("embeddings")

val pipeline = new Pipeline().setStages(Array(documentAssembler, tokenizer, embeddings))

val data = Seq("Ich liebe Funken NLP").toDF("text")

val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("de.embed.bert_base_german_uncased").predict("""Ich liebe Funken NLP""")

Model Information

Model Name: bert_embeddings_bert_base_german_uncased
Compatibility: Spark NLP 3.4.2+
License: Open Source
Edition: Official
Input Labels: [sentence, token]
Output Labels: [bert]
Language: de
Size: 412.8 MB
Case sensitive: true

References

  • https://huggingface.co/dbmdz/bert-base-german-uncased
  • https://deepset.ai/german-bert
  • https://deepset.ai/
  • https://spacy.io/
  • https://github.com/allenai/scibert
  • https://github.com/stefan-it/fine-tuned-berts-seq
  • https://github.com/dbmdz/berts/issues/new