Using 2 GPU on kaggle

[akadir (Abdulkadir Guner)]

Hi,
I tried:
from fastai.accelerate import *

on Kaggle 2 x T4 environment with error:
ModuleNotFoundError: No module named ‘fastai.accelerate’.

fastai version is: 2.7.10.
what am I missing?

[misza222 (Michal Pawlowski)]

@akadir

fastai.accelerate

Hi Abdulkadir,

Did you follow Notebook distributed training – fastai and running on multiple GPUs doesn’t work on kaggle?

[muellerzr (Zachary Mueller)]

You should try and follow the kaggle kernel I made showcasing how to use it Multi-GPU and Accelerate | Kaggle

[jeremy (Jeremy Howard)]

@muellerzr sorry if this is a dumb question, but I’ve been wondering for a while: what’s the benefit of using notebook_launcher as you show there, vs using nbdev to export a script and running it using ! or similar in the notebook after export?