Please post below if you have any questions, need help, or have tips on running the notebook on any cloud notebook platform. For help setting up on your own machine, please post to.
This thread is for discussing questions and tips related to running Jupyter Notebook on cloud platforms. If you have any questions or need help with running Jupyter Notebook on your own machine, please post in this topic.
(Note that Colab has its own topic.)
You don’t need to install CUDA. If you install pytorch with conda, then CUDA will be installed for you automatically.
hello. i followed the first colab setup steps to connect to google drive but nothing is showing there. what am i doing wrong?
Can you share your screen shot and which step you are referring in the documentation?
Hi team,
I have been using Is it a bird? Creating a model from your own data | Kaggle? as my guidance and my Kaggle account to practice.
I think the example in the book is out of date!
or should I use one of the platforms mentioned here to do my practice code?
I am just a bit confused 
Hi @houman.kargaran,
No, you should be able to go with that Kaggle notebook. Did you enable Internet connection? Otherwise, it should work. (I did execute all cells successfully.)
You can see some message from pip complaining about packages, but you can ignore it, as pip quite often complains about incompatible metadata. However, these restrictions are usually not enforced.
If you’re talking about the Deep Learning for Coders with Fastai and PyTorch book, then yes, I guess its content is a bit different from what you can find in the first lesson’s notebook. Like cnn_learner instead of vision_learner and some other minor aspects, like using a different dataset. But I believe that these changes are rather minor. I haven’t yet read through the book, so maybe the next chapters a bit more different. However, in general, the fastai library looks and feels rather similar to what I remember since the last time I tried it out, and it was a few years ago. So I don’t think that examples are too outdated anyway.
Just quick info about Google Colab and performance I’m not up to date with collab but 2 years ago.
google collab was assigning GPU from the available GPU pool so you could be lucky and sometimes get a better or worst assignment. The way to verify it was to get SSH connection to your collab session and execute on it nvidia-smi there were a pretty lot of nice tricks around it 
Sorry about the lack of deatails
Ok even without colab SSH you can always run in colab cell
!nvidia-smi
Thu Apr 28 23:02:35 2022
±----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Tesla K80 Off | 00000000:00:04.0 Off | 0 |
| N/A 33C P8 30W / 149W | 0MiB / 11441MiB | 0% Default |
| | | N/A |
±------------------------------±---------------------±---------------------+
±----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
±----------------------------------------------------------------------------+
Which give you GPU and memory assignment
I was looking at the Kaggle distracted driver competition, at the end of the notebook 01_intro.
The independent variables are pictures of drivers at the wheel of a car, and the dependent variables are categories such as texting.
1.- The predictions are calculated from the independent variable
2.- targets or the dependent variable
and I was wondering how you will construct your data loader? or what kind of data loader it needs, for data and images together using FastAI.
Merging image, tabular
Any suggestions ?
Thanks,
Has anybody else found that Jupyter’s tab complete and shift + tab don’t work for them in the Fast.ai box on Gradient?
By default Gradient uses their own jupyter IDE where nothing quite works the same, unfortunately. However on the left-hand icon bar, at the bottom, there’s a button which switches over to genuine JupyterLab. If you click that, it should all work fine.
You did everything correctly (see the green ticks and run time next to each cell). I could run search_images_ddg afterwards. That means the import was successful.
The statement Unsupported Cell Type. Double-Click to inspect/edit the content. is a separate cell NOT an error message!!! If you double click it, you will see the content as highlighted. Since the fastbook is written in Juypter Notebook, the cell may not be compatible with Google Colab.
@Jeremy please consider mentioning this incident in the next lesson.
Quick question on following the book using the Paperspace. As you can see in the screenshot below I came across a handful of placeholders (<>) in the text but they seem empty.
When you click on that cell to put it in edit mode, you’ll see there’s actually an id in there, which matches to ids used for chapters, figures, etc elsewhere in the book. It’s used for creating cross-references in the paper book, but isn’t supported by Jupyter, so it’s blank.
Hi,
I attempted running the course notebooks on my laptop. For this I conda installed fastai with python 3.7. Further i also conda installed fastbook. However I am not able to run the notebooks an get the error:
ModuleNotFoundError: No module named 'torch''
Doesn’t fastai module install pytorch as well?
In pip it install pytorch along with fastai module. In conda also that should be the case
try following in colab:
from google.colab import drive
drive.mount("/content/gdrive")
Then it should open a new page to ask for login to ur google account