Rasa training on GPU

Rasa Open Source
1

Hello :wave:, Iā€™m currently training my rasa model on cpu and it takes >18hrs to train the model. Rasa version is 3.6. Can some one suggest a best combination of Rasa, tensorflow, cuda and cudnn versions so that I could get my model trained quickly :smiling_face_with_tear: on GPU. My laptop specifications: Windows 11 Rtx 3050 6GB

2 Likes
2

Your NLU model seems to be quite defined and large. If i were you, I would totally separate NLU from rasa and train/host/deploy it somewhere else and integrate it with rasa by defining it in nlu: section at endpoints.