[Rasa NLU] how to train the data with GPU

I use Rasa NLU: 0.13.2. My training data is very large which needs a long time to train. I’m wondering if it is possible to train the data via GPU? If so, can anyone tell me how to do that?

How big is your training data, and how long is long? Also, which pipeline are you using?

@andy51002000 and @akelad im trying to train an x-yml file dataset with 220 MB but it takes too long. Which is the best way to deal with this? By the way it is possible to send this y-xml file in zip format? Thanks

What yaml file? Is this in the rasa training data format? Also 220MB seems like way too much text for a bot

Yes is in the right format, because when i try it with a smaller data set it works fine. But with 220 MB took a lot to load and after 30 min give error.


I already try to reduce the dataset using lookup tables but for example if i search by “jjj”, rasa nlu extract as “A0N_channel” entity but “jjj” is not even in the dataset.

“lookup_tables”:[{“name”:“24KITC, 24KTHD”,“elements”:[“24Kitchen”,“24KITC”,“24Kitchen HD”,“24KTHD”]},{“name”:“ABOLA”,“elements”:[“A Bola TV”,“ABOLA”]},{“name”:“AMC, AMCHD”,“elements”:[“AMC”,“AMC”,“AMC HD”,“AMCHD”]},{“name”:“ARTV”, elements":[“ARTV”,“ARTV”]},{“name”:“AXN, AXNHD”,“elements”:[“AXN”,“AXN”,“AXN HD”,“AXNHD”]},{“name”:“AXNBL, AXBHD”,“elements”:[“AXN Black”,“AXNBL”,“AXN Black HD”,“AXBHD”]},{“name”:“AXNWH, AXWHD”,“elements”:[“AXN White”,“AXNWH”,“AXN White HD”,“AXWHD”]},{“name”:“AFRO”,“elements”:[“Afro Music”,“AFRO”]},{“name”:“ALJAZ”,“elements”:[“Aljazeera”,“ALJAZ”]},{“name”:“ANGELSD, ANGELUS”,“elements”:[“Angelus TV”,“ANGELSD”,“Angelus TV HD”,“ANGELUS”]},{“name”:“BBC E”,“elements”:[“BBC Entertainment”,“BBC E”]},{“name”:“BBC W”,“elements”:[“BBC World”,“BBC W”]}

i’m afraid there’s not much we can do here, i don’t think a 220MB file is necessary for training rasa. How many intents do you have? And how many examples per intent?