I wanted to use GCP Cloud TPU to run my Fastai v2 segmentation model. Is there any tutorial available on how this can be done? I found links on the forum on how to use GPU (https://course.fast.ai/start_gcp.html), but couldn’t find any guide on how to use TPUs. I am new to this, so any help is welcome.
There is no TPU support currently.
Thanks Zachary. Is there no TPU support only for Fast AI v2 or is it also not available for FastAI V1?
If it is available for FastAIv1, then I can change my code to use FastAI v1. If it is not available for both v1 and v2, then I can only use GPU on GCP, right?
Sorry there is some TPU support in v1. Its not great, but it’s there. There is none in v2 currently
Some stuff like vision, tabular and collab filter works but some vision stuff like batch transforms and rnns for NLP don’t.
@ilovescience is I think working on running it on multiple tpu cores.
If you are interested our project is on github.
Thanks a lot. I will have a look at Github.