CRFEntityExtractor or DIETClassifier splits one entity into multiple words

Rasa Open Source
1

I updated from RASA 1.10.2 to 1.10.14.

My pipeline is -:

language: en
pipeline:
  - name: SpacyNLP
    model: en_core_web_md
  - name: ConveRTTokenizer
  - name: ConveRTFeaturizer
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: CRFEntityExtractor
  - name: DIETClassifier
    epochs: 50
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
  - name: SpacyEntityExtractor
  - name: "DucklingHTTPExtractor"
    url: "http://0.0.0.0:8000"
    locale: "en_GB"
    timezone: "US/Pacific"
    timeout : 3

In RASA 1.10.2 CRFEntityExtractor was working fine but in 1.10.14 it is splitting the entity into multiple tokens.

Pls help

If more information for nlu data is required will provide.

Thanks!

2

Looking for a help @RASA Team.

3

check this post: Introducing DIET: state-of-the-art architecture that outperforms fine-tuning BERT and is 6X faster to train

4

I think there’s an issue by using spacy together with convert. They each come with their own tokenizer and I can imagine that they’re not playing nice with eachother. I also wonder, is there a particular reason you’re using the spaCy entity extractor? Are there specific entities that are pretrained in en_core_web_md that you’re interested in?