Looking at the HTTP API documentation for version 1.1.4, in the train a RASA model section and example request body schema example, the training data was in markdown format. Can training data in JSON format be used within the application/json request body? Or does it have to be markdown?
Yes, it can be in json format. Just add the file “nlu.json” in the data folder. The training data from both the files “nlu.json” and “nlu.md” will be used for training if it exists.
Thank you for the response, but can the json training data be provided inline similar to how to markdown nlu training data is provided in the example?
@mscott12 I think you can’t. Checkout tracy to generate the files in json
format. Also, checkout Convert a Rasa NLU training file from JSON to Markdown format · GitHub to convert json
to md
format.
I just looked at the server file. Doenst seem like it allows for json files as this line joins the temp dir with nlu.md which is then used in the “training_files” arg for the actual training Maybe we can add a little flag saying hey this is json? the training files arg supports json, just not the rasa http endpoint.
@KarthiAru can you say me bit in detail… What are the other requirements to train json file like what about stories.md file?