How can I make "rasa train" take advantage of GPU on local machine

Rasa Open Source
1

I have a NVIDIA Quadro P520 on my laptop and I would like to take advantage of that when training my rasa model (rasa train). How can I enable GPU/CUDA for an existing rasa project? I am using 2.5.0

2

Rasa should automatically use your GPU during training. You should be able to check that your GPU is visible and that it is handling the job by using the nvidia-smi tool from the command line. https://developer.download.nvidia.com/compute/DCGM/docs/nvidia-smi-367.38.pdf

This will show your GPU if available and then during training, you’ll see the memory utilization increase.

3

This blog post may be helpful.

4

Yeah I did the nvidia-smi while training the rasa model and it only showed my available GPU but no utilization at all. So for some reason, Rasa is not automatically training on GPU while it is available.

1 Like
5

I’m having the same problem on an Ubuntu VM in GCP.

6

I have same problem @baurjanj .Whether works Please let me know how you done if so worked.

7

Similar discussion: GPU utilization is always 0

8

Hi guys, I have the same problem, did you manage to solve it? @baurjanj