Run model on a CPU?

[nicolagheza]

Hello, would it be possible to run the trained model on a CPU?
I download the 'demo model' from dropbox but when I run the inference.py from my laptop I get:
ImportError: libcublas.so.8.0: cannot open shared object file: No such file or directory`

Does the model need a GPU also in production?

[daniel-kukiela]

Yes, you can run it on CPU.

pip uninstall tensorflow-gpu
pip install tensorflow

Replace pip with whatever you are using, probably pip3 or pip3.6

[nicolagheza]

Thanks, it worked.
But I had to use tensorflow==1.4 to make it run properly.
With 1.7 I had some errors.

[daniel-kukiela]

Can you tell me more about those errors? I think i might know what your issues are.
Also can you tell me what CPU are you using (model of CPU)?

[nicolagheza]

Hello,
My processor is Intel® Core™ i7-6600U CPU @ 2.60GHz × 4
I actually cannot tell you more about those errors because now It is working also with Tensorflow-1.7
(I am using Ubuntu 16.04)
The inference script now works properly using your model. Is it normal that I don't get proper answers right?
Like this is the answer to "How are you doing?":
"Yo^o^o^o^o^o^o^o^o^o^o^o^e^o^o"

[daniel-kukiela]

So the problem was not what i was thinking it is.

As for responses - it's pretty normal at the training begionning. How many training pairs and how log model is training now?

[nicolagheza]

What do you mean by that? Log model? I have been using the pre-trained model to test it out. I did not train it myseld. I am trying to get it to train on Google ML engine because I don't have a GPU. Can I just train it starting from your pre-trained model and the available data?

…

On Thu, 12 Apr 2018, 23:17 Daniel Kukieła, ***@***.***> wrote: So the problem was not what i was thinking it is. As for responses - it's pretty normal at the training begionning. How many training pairs and how log model is training now? — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#58 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABG-3QlJ_4RKwl7YtvBkzmTHB_AyVLzoks5tn8RigaJpZM4TO0Yd> .

[daniel-kukiela]

Oh, so it's a very simple model trained quickly, as i didn't have resources for a longer training.
You can train more, you need to prepare your training set ang go through all steps to prepare and train your model. The reason for that is preparation steps makes files (training data and vocab files) for training set you are using.