Russian DistilBERT Embeddings (from Geotrend)

Description

Pretrained DistilBERT Embeddings model, uploaded to Hugging Face, adapted and imported into Spark NLP. distilbert-base-ru-cased is a Russian model orginally trained by Geotrend.

Download Copy S3 URI

How to use

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

tokenizer = Tokenizer() \
.setInputCols("document") \
.setOutputCol("token")

embeddings = DistilBertEmbeddings.pretrained("distilbert_embeddings_distilbert_base_ru_cased","ru") \
.setInputCols(["document", "token"]) \
.setOutputCol("embeddings")

pipeline = Pipeline(stages=[documentAssembler, tokenizer, embeddings])

data = spark.createDataFrame([["Я люблю искра 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 = DistilBertEmbeddings.pretrained("distilbert_embeddings_distilbert_base_ru_cased","ru") 
.setInputCols(Array("document", "token")) 
.setOutputCol("embeddings")

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

val data = Seq("Я люблю искра NLP").toDF("text")

val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("ru.embed.distilbert_base_cased").predict("""Я люблю искра NLP""")

Model Information

Model Name: distilbert_embeddings_distilbert_base_ru_cased
Compatibility: Spark NLP 3.4.2+
License: Open Source
Edition: Official
Input Labels: [sentence, token]
Output Labels: [bert]
Language: ru
Size: 202.3 MB
Case sensitive: true

References

  • https://huggingface.co/Geotrend/distilbert-base-ru-cased
  • https://www.aclweb.org/anthology/2020.sustainlp-1.16.pdf
  • https://github.com/Geotrend-research/smaller-transformers