Hi, I’m a beginner on this fastai world, but before I invest more time learning, I was hoping to get a bit more clarity on the architecture side to see if it will work on the current structure I have at the company I work.
I understand that to train models we need a GPU and I’m planning to use one of the cloud options to do so. However, once the model is already trained, I would like to embed into the current libraries we use at our company. Currently, the python kernel we use is running on windows without a decent GPU. Good to mention that the models I’m planning to train are classification models on structured data, so no images or video. I was wondering if there is a way to run a fastai model locally on a windows machine without a decent GPU.
You Can run your model without GPU. If your computer does not have any it will run on CPU.
You only cannot use
learn.fp16() which requires GPU. The network will return results 5-50 times slower.
Fastai2 does not support windows yet (checkout https://forums.fast.ai/t/fastai2-on-windows/)
but if you have windows 10 you can use WSL2 (which is Linux-on-Windows), or Virtualbox or Docker
It will work there, but if you want to use GPU latter its not trivial to configure Nvidia on any of these. (I’m not sure if it is possible, I couldnt make it work)
Hi, I’ve managed to install fast ai v2 on my personal win10 with GPU using conda and I’m training models there. Is win10 officially supported now? Is it possible to install fast ai on a win10 machine without a cuda gpu just to run predictions?