to follow up to my prior issue, looks like it was a self-inflicted anaconda wound! After searching around for “jupyter kernel dead won’t restart” I landed on:
Apparently after multiple installs of anaconda you end up with a few kernel.json files (check the two highlighted)
when I compare the paths inside, the one in \envs\fastai… had the path: C:/ProgramData/Anaconda3//envs/fastai/bin/python
while the main one in \share\jupyter… had: C:\\ProgramData\\Anaconda3\\python.exe
After replacing the path from the fastai kernels.json with the one that contained the .exe , voila! Now the notebook successfully loads with a ready kernel!
Albeit another error on not finding spacy when trying to import
---------------------------------------------------------------------------
ModuleNotFoundError Traceback (most recent call last)
<ipython-input-2-c51a52575a56> in <module>()
1 # This file contains all the main external libs we'll use
----> 2 from fastai.imports import *
~\Documents\GitHub\fastai\courses\dl1\fastai\__init__.py in <module>()
----> 1 from .basic_train import *
2 from .callback import *
3 from .callbacks import *
4 from .core import *
5 from .data import *
~\Documents\GitHub\fastai\courses\dl1\fastai\basic_train.py in <module>()
1 "Provides basic training and validation with `Learner`"
----> 2 from .torch_core import *
3 from .data import *
4 from .callback import *
5
~\Documents\GitHub\fastai\courses\dl1\fastai\torch_core.py in <module>()
1 "Utility functions to help deal with tensors"
----> 2 from .imports.torch import *
3 from .core import *
4
5 AffineMatrix = Tensor
~\Documents\GitHub\fastai\courses\dl1\fastai\imports\__init__.py in <module>()
----> 1 from .core import *
2 from .torch import *
~\Documents\GitHub\fastai\courses\dl1\fastai\imports\core.py in <module>()
4 import abc, collections, hashlib, itertools, json, operator
5 import mimetypes, inspect, typing, functools
----> 6 import html, re, spacy, requests
7
8 from abc import abstractmethod, abstractproperty
ModuleNotFoundError: No module named 'spacy'
I tried to conda install -n fastai spacy and ran successfully, but still got the same error log
check the contents of your old\fastai folder vs the one in the repo. mine was somehow different, so i had to replace it manually with the correct one from the repo.
Anyone had issues with pytorch 1.0 finding CUDA on windows?
I’m trying to get fastai 1.0 up in windows, with cuda 9.0 installed. I’ve built pytorch and created a new conda env with py3.7 and:
Python 3.7.0 (default, Jun 28 2018, 08:04:48) [MSC v.1912 64 bit (AMD64)] :: Anaconda, Inc. on win32
Type “help”, “copyright”, “credits” or “license” for more information.
import torch
print("PyTorch version: ", torch.version )
PyTorch version: 1.0.0a0+0e966fc
print("CUDA version: ", torch.version.cuda) CUDA version: None
print(“CUDA available”, torch.cuda.is_available()) CUDA available False
print("CUDNN enabled: ", torch.backends.cudnn.enabled)
CUDNN enabled: True
previous environment still finds cuda
Python 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 10:22:32) [MSC v.1900 64 bit (AMD64)] on win32
Type “help”, “copyright”, “credits” or “license” for more information.
import torch
print("PyTorch version: ", torch.version )
PyTorch version: 0.3.0b0+591e73e
print("CUDA version: ", torch.version.cuda) CUDA version: 9.0
print("CUDNN enabled: ", torch.backends.cudnn.enabled)
CUDNN enabled: True
(deja vu… months back in this thread when i first installed fastai, cuda wasn’t found until after a couple reboots. - admittedly I’ve only tried one reboot this time)
still having issues with dead kernel when trying to run notebooks.
Reinstalled anaconda and reinstalled fastai from github. Cleared conda cache. Ran conda env update and then manually had to add pyproj and shapely using conda install -n fastai pkgname.
after looking up jupyter kernelspec list appears I have two (depending if I activate fastai or not):
The contents of kernel.json @ C:\Users\Stephen\Anaconda3\share\jupyter\kernels\python3 are…
Can anyone kindly help me with the installation of fast ai on windows 10. I dont have access to conda and git environment due to network restrictions.So im not able to install Fast ai libraries.so i used pip install and it gave the below error.
You can try two things.
Copy the folder fastai to the folder where the notebook page is located.
or
include on top of your notebooks at the very beginning before the fastai stuff
import os
import sys
sys.path.append(“PATH TO THE FASTAI FOLDER”)
I followed all steps from jeremys original post, however, when I run “lesson1.ipynb”, I receive the following error in line [3] from fastai.transforms import *
"AttributeError: module ‘torch’ has no attribute ‘float32’ "
I managed to run lesson1.ipynb on my laptop with NVIDIA GeForce 840M GPU with Windows 10 following your instructions. But the PyTorch version which got installed conda env update did not support my GPU. I got this error message:
“Found GPU0 GeForce 840M which is of cuda capability 5.0. PyTorch no longer supports this GPU because it is too old.”
After scavenging the forums, I followed the instructions listed on this thread:
Finally, everything worked. But when I monitored CPU and GPU performance on Task Manager while learning, the learner’s GPU utilisation was around 2.5-3 percent whereas the CPU utilisation was much higher around 50 percent (in certain stages).
I wonder if this is normal. Thanks. In any case, being able to do deep learning on my own laptop feels nice.
During environment setup I got the following errors:
pexpect 4.6.0 requires ptyprocess>=0.5, which is not installed.
spacy 2.0.16 has requirement regex==2018.01.10, but you’ll have regex 2018.8.29 which is incompatible.
I’m on Windows 10 using Anaconda conda 4.5.11 python 3.6.6
After redoing the symlink to fastai I got errors about missing Module spacy
I then ran
conda install -c conda-forge spacy
some packages were downgraded and then I got missing Module dataclasses
after
conda install -c conda-forge dataclasses
some packages were downgraded
now I’ve got missing Moduie fastprogress
I can’t get PyTorch to use my GPU. It says that my graphics card is too old (Gtx 760). It says that PyTorch 0.3.0 will work but I can’t find it to install. Anyone find it? with instructions?