Different results from rasa shell nlu on different machines

Hi I have exactly the same nlu, rules, stories and domain and config.yml on two different machines. I have deleted the old models and I use rasa train --force on both and I get different results. On one I get this result whenI run rasa shell nlu { “text”: “set filter to farm,peak”, “intent”: { “name”: “set_category”, “confidence”: 0.8906326293945312 }, “entities”: [ { “entity”: “category”, “start”: 18, “end”: 19, “confidence_entity”: 0.9727645516395569, “value”: “,”, “extractor”: “DIETClassifier” }

on the other I get “text”: “set filter to farm,peak”, “intent”: { “name”: “set_category”, “confidence”: 0.8906326293945312 }, “entities”: [ { “entity”: “category”, “start”: 18, “end”: 19, “confidence_entity”: 0.9727645516395569, “value”: “,”, “extractor”: “DIETClassifier” } Diiferent confidence, but I can accept that, but not that the value is different depending on which machine I am on. What am I doing wrong? Is it using something from the cache in one case but not in the other when I train the model? Maybe I should move the model from one machine to the other to see if there is a difference?

Hope you have some good answers!!! Mats

The value of the entity category is , in both cases. No difference.