Yup. Any thoughts on how can I pass a path to my learn.export() object?
I understand. Thanks for the detailed reply. Are there any possible workarounds to this ?
you can simply set learn.path to any directory you want where it will then save your model.
Simply: learn.path = XYZ
learn.path = Path('notebooks/course-v4/nbs')
This should fix it.
Hi @dkobran and @tomg . I am using a paid subscription to paper space and I am getting an error when running the following block of code in Notebook 10.
dlf_lm = DataBlock(
blocks = TextBlock.from_folder(path, is_lm = True),
get_items = get_imdb, spliter =
RandomSplitter(0.1)).dataloaders(path,path=path,bs=128,seq_len=80)
I also encountered a similar error I posted about here
Can you please advise?
This looks like a question for @sgugger. /storage/data is where the public datasets consumed by your fastai notebooks live. I’m not aware of any dataset called ‘imdb_tok’, just imdb. Sylvain, can you shed some light on this?
@akashpalrecha Sorry about the delay. You could just run a sidecar Job (or Core VM) though I think the best option is probably to launch TensorBoard directly in a notebook:
https://medium.com/hackernoon/how-to-run-tensorboard-for-pytorch-1-1-0-inside-jupyter-notebook-cf6232498a8d. I have never done this but it looks pretty straightforward. If you try it out, let us know how it goes!
imdb_tok is autogenerated by the tokenizer for the imdb set
As per the above, this looks like the output was never generated.
The storage/data/ does not have write access right? Or does it?
@tomg Any suggestions on how I should proceed? It seems to me link I will face this issue any time I am using PaperSpace for NLP related tasks.
Hi, this doesn’t appear to be a Paperspace specific issue. My fastai notebook generated two pkl files without issue.
:/notebooks# ll storage/data/imdb_tok/
total 8040
drwxr-xr-x 7 root root 4096 Mar 18 05:34 ./
drwxr-xr-x 3 root root 4096 May 23 20:19 ../
-rw-r--r-- 1 root root 3649850 Mar 18 05:34 counter.pkl
-rw-r--r-- 1 root root 3070555 Mar 18 05:34 lengths.pkl
drwxr-xr-x 4 root root 4096 Mar 18 05:26 test/
drwxr-xr-x 2 root root 4096 Mar 18 05:26 tmp_clas/
drwxr-xr-x 2 root root 4096 Mar 18 05:26 tmp_lm/
drwxr-xr-x 4 root root 4096 Mar 18 05:26 train/
drwxr-xr-x 2 root root 1478656 Mar 18 05:34 unsup/
To clarify, /storage/data is fully writeable, just the specific preopulated dateset directories are not (for eg. storage/data/imdb). This seems that whatever operation you are running is failing to properly complete – for help with that I ask you direct your question to the fastai course proctors.
Hope this helps,
–Tom
It has full write access. Only the specific dataset sub-directories are read-only, such as storage/data/imdb, etc.
I tried running that block of code after deleting the imdb_tok folder. That seems to have done the trick.
4 Likes
Okay, then it’s probably you had run it once with an older version of fastai2 and it didn’t have all the expected files I’d guess.
Hello. How to update fastai2 in Paperspace?
I’m not talking about running pip install --upgrade fastai2 in a terminal in Paperspace.
Why? Because the pip version of fastai2 is not regularly updated (which is normal).
In an ubuntu installation, it is easy: git pull in the fastai2 folder 
In Paperspace?
Thanks for the tip!
I had earlier tried it before posting on the forums but it gave me the same nginx server error.
It’s weird that there seems to be no way to run tensorboard on paperspace gradient notebooks.
Meanwhile I’m switching to wandb for my logging needs!
Is it just me or does it take forever for a notebook on paperspace to open up? I typically have to wait 10 -15 mins when it used to be instantaneous.
1 Like