@brismith I am not able to import the modules such
fastai.torch_imports , fastai.io, fastai.model, fastai.dataset I couldn’t find torch_imports and fastai.io even in the github repo. Any ideas how i can fix these. Even on kaggle, these modules seem to be missing
What version of FastAI do you have? I think this was a recent change - I have 1.0.39 and these import just fine.
Try:
conda update -c fastai fastai
and see if that solves the problem.
These were my steps - since then I have updated to FastAI 1.0.39. Some of us have noticed that Windows is slow getting going at the start and between epochs when training.
https://forums.fast.ai/t/pytorch-1-0-is-officially-released/32212/15?u=brismith
I packed it in after a couple days trying to get it to work. I don’t think Fastai is quite ready for public consumption yet!
However; the course is excellent I am watching the lectures and then completing the exercises using other tools. (actually probably a better learning experience in the long run)…
hi, I am trying to install fasti on a windows machine running cuda 10. I am facing a lot if difficulty also i dont seem to be able to access your post. Is it possible you can re-post it somewhere that I can access.
regards,
Here is the posting @sahand68. I think now the Conda Kernels may be in the latest FastAI.
I’m getting past that error with Python 3.7 and FastAI 1.0.37.
This is the steps I’m using for my Conda environment. Windows 10.
conda create -n fastai-3.7 python=3.7
conda activate fastai-3.7
conda install pytorch torchvision -c pytorch
conda install -c fastai fastai
conda install nb_conda_kernels
python -m ipykernel install --user --name fastai-3.7 --display-name “Python (fastai-3.7)”
conda install ipywidgets
2 Likes
I tried your approach on windows 10 using latest Fastai and python 3.7. I successfully run “from fastai.imports import *” .But I still got the following error. Do you know how can I fix this?
ModuleNotFoundError Traceback (most recent call last)
in
----> 1 from fastai.transforms import *
2 from fastai.conv_learner import *
3 from fastai.model import *
4 from fastai.dataset import *
5 from fastai.sgdr import *
ModuleNotFoundError: No module named ‘fastai.transforms’
I’ve seen a few people have similar issues @saeedm and it looks like you are not getting the latest version of fastai (1.0.40 I think). The different import statements came in around 1.0.39 or so. It might be best to start clean and delete Conda and start again - to be sure it pulls the right version. To see what you have Conda list may show fastai 1.0.32 - or almost certainly a number lower than 39.
Hi all. I have run into this problem and have read ans followed the instructions of the first post. but still getting this error.
i deleted the fastai myself therefore del fastai shows that.
please help. spent half the day on it x |
mine shows 1.0.38. going start from beginning X-(
Hey, Is it possible to install fastai with CUDA v8.0 already installed on a windows system? If I change the CUDA version to 8.0 in the environment.yml, would there be any issue installing the fastai and using it later? Please advice.
Another doubt is could we use Virtualenv instead of Anaconda?
Thanks,
Siddhartha
I could install fastai on Windows 10 after following Jeremy’s instructions. But then I started getting errors saying that my GPU is old. Tried multiple fixes including installing older versions of pytorch but did not work. Finally gave up and started using crestle.com as per Jeremy’s instructions. Last week I switched to Google Colab and am using that now. It requires a little bit of setup but is worth it. I switched to Colab because I am a slow learner and the cost was piling up in Crestle.
What I learnt is that I wasted too much time trying to get it to work on my laptop instead of actually doing the lessons. Murphy’s Law “If everything else fails, follow the instructions”.
Hi, I’m new here and trying to setup my own laptop to get started with the courses. I did all the steps from step 1 down to step 7, then step 8 in order to get ready for the DL course. After that, I typed jupyter notebook and notebook opens, and then I open the “lesson1.ipynb” notebook. Afterward, I went down to get to the first runnable section, ie. the loading section. The first section works well, but the second section is constantly giving me an error “DLL load failed: The specified module could not be found’”.
Is there anything I’m probably missing?
Thanks in advance
If your python version is 3.7.2 or 3.6.8, a fix has just been released. Please try to update python.
conda update python
The howto at the top worked flawlessly for me. The only glich I’ve encountered is that I didn’t start by putting the Data and the Weights folders in \DL1. It was a bit baffling to see and be able to open the data folder in the notebook, but get a ‘file not found’ when running dogbreed.
I bought a refurbished Dell 3010 from nvblindchildren.org for $150 and got a GeForce GT 1030 card for $85. The only risky decision was to ignore the 300 watt power supply minimum called for by the card maker.
Please, no sneering at my wimpy rig. It works, it is always available AND I can leave it on for days or weeks and not worry about hourly charges.
I’m not sure whether to include the cost of the KVM, but it makes it easy to keep working on other stuff while I wait for epochs to run. It’s a TesMark HDMI 4 port, cost: $120. I had planned to get it anyway.
I’m confident the rig is good, because I get a true response when I run
#import torch
#torch.cuda.is_available()
however Task Manager shows 100% sometimes for the CPU and the highest I’ve seen on the GPU was 10%. Are there any better tools for monitoring GPU utilization?
I came across a comment somewhere that if the GPU is installed correctly, it’s really up to proper application of the driving software to get maximum utilization. My tech skills are stone age imho. I don’t know where to look, but I’ve got a hunch that I’m not getting the most I could out of the GPU. Then again, I don’t want to melt a hole in the motherboard.
Off topic comments follow:
I realize it would be far easier on Jeremy if he weren’t targeting Windows as a platform on which the training can be conducted. Its a real distraction from the enormous effort that goes into conduction the classes, guiding actual students and keeping the course work up to date.
I hereby send a big THANK YOU to him and his staff. Please keep up the brilliant work.
In the installation instructions on this post there is a detour to install a jpeg decoder:
conda uninstall --force jpeg libtiff -y
conda install -c conda-forge libjpeg-turbo
CC="cc -mavx2" pip install --no-cache-dir -U --force-reinstall --no-binary :all: --compile pillow-simd
I was able to run the first 2 lines on my 64-bit windows 10 machine; is there a windows equivalent for the 3rd line? If not how do I unistall libjpeg-turbo and revert o libtiff?
I know it’s not elegant, but can you just install libtiff on top? Pointers should be updated with the install, so it won’t be looking for libjpeg-turbo.
This has it’s risks, but if you are using a force process, I assume you have a boot disk that works and system backup/recovery file that is up to date.
Thanks, @mike00. I actually don’t have a boot disk that works or backup/recovery file.
I suppose that another workaround would be to uninstall and re-install anaconda, since I just installed it yesterday.
If that’s the case, then yes reinstalling Anaconda is a viable choice.
I didn’t mean to alarm you with the talk of backup, but I’ve saved myself an awful lot of time by getting a stable build and taking a backup at that point. Then, no matter how much I mess it up, it doesn’t take too much time to recover. I’m an old systems guy and habits die hard.
What problem were you trying to solve with the jpeg turbo thing?
No particular problem. I am installing fastai on my new machine and I was trying to gather up a new shiny object 