Reloading a model in a running system using rasa_nlu.train

Rasa Open Source
1

Hello

Im trying to write a script that collects .md files, combines them and retrains an existing model. If I use the Web API erverything works fine (using /train).

But if I try to use:

python -m rasa_nlu.train -v --config ${CONFIG} --data tmp.md --path projects --project ${PROJECT} --fixed_model_name ${MODELNAME}

it only creates a model but the server wont reload it.

Is there any way to trigger this process ? Did I miss something in the docs ?

Version: 0.13.1-full (Docker image)

config:

language: "de"

pipeline:
- name: "tokenizer_whitespace"
- name: "intent_entity_featurizer_regex"
- name: "ner_crf"
- name: "ner_synonyms"
- name: "intent_featurizer_count_vectors"
- name: "intent_classifier_tensorflow_embedding"

Thank you in advance

see: stackoverflow

2

If you want the server to load it, you’d have to restart the server. Or do it with the /trainendpoint. Also what do you mean retrains an existing model?

3

If I call host:5000/train?model=${MODEL}&project=${PROJECT} it seems to replace the existing model. Thats what i expected when i used the python based approach.

4

Are you using a new model name or an already existing one? Because yes it will overwrite if it already exists

5

Yes the model already exists but when using the python approach it wont load the overriden model.