Description
Pretrained CamemBertForQuestionAnswering model, adapted from Hugging Face and curated to provide scalability and production-readiness using Spark NLP. camembert_base_qa_fquad is a French model originally fine-tuned on a combo of three French Q&A datasets:
- PIAFv1.1
- FQuADv1.0
- SQuAD-FR (SQuAD automatically translated to French)
Predicted Entities
How to use
Document_Assembler = MultiDocumentAssembler()\
.setInputCols(["question", "context"])\
.setOutputCols(["document_question", "document_context"])
Question_Answering = CamemBertForQuestionAnswering("camembert_base_qa_fquad","fr")\
.setInputCols(["document_question", "document_context"])\
.setOutputCol("answer")\
.setCaseSensitive(True)
pipeline = Pipeline(stages=[Document_Assembler, Question_Answering])
data = spark.createDataFrame([["Où est-ce que je vis?","Mon nom est Wolfgang et je vis à Berlin."]]).toDF("question", "context")
result = pipeline.fit(data).transform(data)
val Document_Assembler = new MultiDocumentAssembler()
.setInputCols(Array("question", "context"))
.setOutputCols(Array("document_question", "document_context"))
val Question_Answering = CamemBertForQuestionAnswering("camembert_base_qa_fquad","fr")
.setInputCols(Array("document_question", "document_context"))
.setOutputCol("answer")
.setCaseSensitive(True)
val pipeline = new Pipeline().setStages(Array(Document_Assembler, Question_Answering))
val data = Seq("Où est-ce que je vis?","Mon nom est Wolfgang et je vis à Berlin.").toDS.toDF("question", "context")
val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("fr.answer_question.camembert.fquad").predict("""Où est-ce que je vis?|||"Mon nom est Wolfgang et je vis à Berlin.""")
Model Information
| Model Name: | camembert_base_qa_fquad |
| Compatibility: | Spark NLP 4.3.0+ |
| License: | Open Source |
| Edition: | Official |
| Input Labels: | [document_question, document_context] |
| Output Labels: | [answer] |
| Language: | fr |
| Size: | 411.9 MB |
References
https://huggingface.co/etalab-ia/camembert-base-squadFR-fquad-piaf
Benchmarking
{"f1": 80.61, "exact_match": 59.54}