Different models running on different ports detecting intents of each other

Hi, I have 2 different models trained on 2 different datasets, these two models are running on 2 different ports 5005 and 5006, but the model running on 5006 detecting the intents of 5005 and vice versa, these are the train command and run command I am using.

rasa train --data data/model1_data nlu.md --out models/model1_models
rasa run --enable-api -m models/mode1_models -p 5006

rasa train --data data/model2_data nlu.md --out models/model2_models
rasa run --enable-api -m models/mode2_models -p 5005

can somebody explain how it works, I want each model to give results of their respective training data?

What domain are you using? It seems like that both models are using the same domain file. Is that correct? In that case both model would know about all intents present in both models.

Hi Tanja, sorry for the late reply, I am using RASA nlu only, I am not using actions, I think there is no need of domain file, now I created 2 separate rasa project and running them separately, It will be very helpful if we can run multiple models in same project on different ports.

Yes, you don’t need a domain file if you are just using Rasa NLU. I tried to reproduce the problem you are describing, but failed. Can you please share the output of the rasa train commands? I’m interested in the training data stats that are logged during training. Thanks.

Hi Tanja, this is my setup,

train log for model1

train log for model2

If you look at the training log output, you will see that both models are trained on the same data - or at least both datasets have the same intents, entities and the same number of training and test examples.

What commands are you executing to train your models? You should use rasa train --data data/data1/nlu.md --out models/model1. If you use the command you shared ealier rasa train --data data/model1_data nlu.md --out models/model1_models, the data path will not be found as it does not exists. Thus, the default data location will be used data. All files in that folder that match the NLU file format will be used for training. So, both nlu files will be used for both models.

1 Like

Thank you Tanja, The intents were different, the problem was with the commands I was using for training.