👋 Introducing the Rasa Playground

Bot is not responding as per the expectation

post by Akanksha1 on Nov 17, 2020

Prediction of nlu is very high on irrelevant data.

NLU Pipeline-

pipeline:

• name: ConveRTTokenizer “model_url”: https://github.com/connorbrinton/polyai-models/releases/download/v1.0/model.tar.gz

• name: ConveRTFeaturizer “model_url”: https://github.com/connorbrinton/polyai-models/releases/download/v1.0/model.tar.gz

• name: CRFEntityExtractor

• name: DIETClassifier entity_recognition: False epochs: 300

Before the PolyAI team has decided to take down the ConveRT models from the public domain, I was using the same pipeline

• name: ConveRTTokenizer
• name: ConveRTFeaturizer
• name: CRFEntityExtractor
• name: DIETClassifier entity_recognition: False epochs: 300

but it was behaving good. After changing model_url to https://github.com/connorbrinton/polyai-models/releases/download/v1.0/model.tar.gz model start behaving badly.

Is there any change or difference in between model files which are hosted on given url (https://github.com/connorbrinton/polyai-models/releases/download/v1.0/model.tar.gz) and original convert model files

post by erohmensing on Nov 20, 2020

Hi @Akanksha1, the link you mentioned from connorbrinton is not endorsed by PolyAI, so it could be that something has changed in the model. You may have better luck with this URL: https://github.com/davidalami/ConveRT/releases/download/1.0/nocontext_tf_model.tar.gz

Which, while not officially published by PolyAI, was publicly endorsed by the author of the model: https://twitter.com/matthen2/status/1326851604109172736?s=20