mpetrikov
(Maksim Petrikov)
March 13, 2021, 10:54pm
1
Fast ai was installed using a command by this link https://docs.fast.ai/:
conda install -c fastai -c pytorch -c anaconda fastai gh anaconda
fastai.version is 2.2.7
I try to do the following tutorial:
There is a problem with the step “learn.fine_tune(1)”
I don’t see any progress
I have been waiting for more than an hour.
My laptop technical characteristics:
MacBook Pro (Retina, 15-inch, Mid 2015)
Processor 2,2 GHz Quad-Core Intel Core i7
Graphics Intel Iris Pro 1536 MB
Memory 16 GB 1600 MHz DDR3
Could it be a problem?
My guess is that since you’re not using nvidia graphics card, it can take a very long time. Try using Google colabs for the job and see if that helps
mpetrikov
(Maksim Petrikov)
March 14, 2021, 5:12pm
3
Thank you, Simon. I think that Colab will be work, but I would prefer to solve this problem on my computer.
The problem in the file: fastai/data/load.py
in this line
for b in _loaders[self.fake_l.num_workers == 0](self.fake_l):
This line freezes.
Loaders:
(<class ‘torch.utils.data.dataloader._MultiProcessingDataLoaderIter’>, <class ‘torch.utils.data.dataloader._SingleProcessDataLoaderIter’>)
I think that the problem in the _MultiProcessingDataLoaderIter loader and it isn’t fastai problem.
muellerzr
(Zachary Mueller)
March 14, 2021, 5:45pm
4
As mentioned, you don’t have a NVIDIA graphics card or a CPU capable of multiple workers. As a result you need to set num_workers to 0 in your call to .dataloaders, and can expect fit to finish in multiple hours if not more (vs just minutes in a Colab or other GPU environment) as you’ll be in a CPU-only environment
1 Like