I’m a newbie to this as well, and am wondering if anyone else is having my problems.
I have windows 10 and am connecting through chrome browser.
I never seem to be able to start my paperspace machine - just spins and spins and I keep going back to the console and trying to start it again.Really frustrating.
Then, when I get it started, it is impossible to open a jupyter notebook - I keep getting a note saying my computer is unable to connect. (Yes, I replaced local host with the location of my virtual machine).
I am really frustrated with this entire process. Shouldn’t it be easier to be part of your course? I spend hours watching things not connect or not able to connect. What am I doing wrong?
Sorry to be such a newbie…Hoping to learn something…
Just followed the directions at https://github.com/reshamas/fastai_deeplearn_part1/blob/master/tools/paperspace.md for creating a Paperspace account. Waited for 15 minutes to receive an email stating that my server was finished provisioning. Then deleted that server and tried again, with same results, that is, server never finished provisioning. The UI now shows the status as inaccessible. Fun.
I am on the fastai paperspace template and having trouble getting tensorflow to recognize/use the GPU. When I train stuff, nvidia-smi shows no usage, because tensorflow is using the CPU.
When I run
import tensorflow as tf
with tf.device('/gpu:0'):
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)
print(c)
I get
Tensor("MatMul:0", shape=(2, 2), dtype=float32, device=/device:GPU:0)
all is good!
but when I run
from tensorflow.python.client import device_lib
device_lib.list_local_devices()
I get
[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 8013515368465626270
]
I googled a bunch and figured out that running
sudo ldconfig
and getting
/sbin/ldconfig.real: /usr/local/cuda-9.1/targets/x86_64-linux/lib/libcudnn.so.7 is not a symbolic link
is very bad news and may require reinstalling cuda stuff (which sounds awful), so I was wondering whether anybody else has run into this before venturing down that deep dark hole. I feel like I could just tell tensorflow to use the GPU without the decorator?
Thanks!
Hi @bryonbean really sorry that the machine is stuck! That should really never happen and I know it is frustrating. Can you send an email to support@paperspace.com so we can get an engineer to look at it ASAP and credit your account back.
What’s not clear is if we use the template, do we do the “Part III Updating fastai repo contents” steps of reshamas script or not. I understand the template replaces the need for the “curl…” step, but do we do the Part III steps, or skip those? Thanks in advance!
Yes you need to do that.
I’ve setup the paperspace fast.ai template and performed git pull and conda env update but I’m still getting this error when trying to run the lesson 1 jupyter notebook. Anyone know how to fix this?
Traceback (most recent call last):
File "/home/paperspace/anaconda3/envs/fastai/lib/python3.6/site-packages/IPython/core/interactiveshell.py", line 2910, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "<ipython-input-3-b4501b23c5cb>", line 1, in <module>
from fastai.transforms import *
File "/home/paperspace/fastai/courses/dl1/fastai/transforms.py", line 2, in <module>
from .layer_optimizer import *
File "/home/paperspace/fastai/courses/dl1/fastai/layer_optimizer.py", line 2, in <module>
from .torch_imports import *
File "/home/paperspace/fastai/courses/dl1/fastai/torch_imports.py", line 19, in <module>
from .models.nasnet import nasnetalarge
File "/home/paperspace/fastai/courses/dl1/fastai/models/nasnet.py", line 620
return model
^
SyntaxError: 'return' outside function
try using the west coast region, that fixed it for me
Hello everyone, I just finished setting up a machine in paperspace and I was making sure everything was working by executing the first lesson code in jupyter but I’m getting this and I don’t know what it is
Any idea?
@lukeharries I am trying to redo the curl http://files.fast.ai/setup/paperspace | bash but using sudo before the command, I’ll let you know how it goes
@teidenzero I’ve tried Crestle which worked but it was so slow, even just importing the libraries, that it was not usable
for anyone else reading this: it has now been fixed. Perform ‘git pull’ in the /fastai folder to pull the fix
Here’s a referral code for $10 - 92GJX60
ATB!
I’m getting the following error:
(fastai) paperspace@psjfhh3rl:~/fastai$ nvidia-smi dmon
Unable to determine the device handle for GPU 0000:00:05.0: Unable to communicate with GPU because it is insufficiently powered.
This may be because not all required external power cables are
attached, or the attached cables are not seated properly.
This happened after the latest git pull. torch.cuda.is_available() returns false. To figure out why, I ran nvidia-smi dmon as shown above. The error indicates no GPU?
I’m using Paperspace with the template AMI.
EDIT: for anyone who ran into this issue, I had to restart the machine.
@jlrw, I had the same problem! Nothing could fix it, also looping in their support didn’t help, and I even couldn’t ssh into the machine.
This resolved it: I deleted the machine running form their Amsterdam center (I’m German), and set one up at the East Coast, after several attempts on the AMS data center failed. The East Coast machine works like a charm, so maybe give setting it up at another center a try!
Note: 1.) I used the ubuntu template - please see comments above: by now, the best option seems to be the fastai template
2.) If you delete a machine, the money paid for storage will only be charged as fraction of a month. So you will not be charged again for storage!
Hope that helps!