deeplearning plugin to work with cuda 9.0

Solved!
longhowlam
Level 3
deeplearning plugin to work with cuda 9.0
Hi all,

is there an easy way to get the deeplearing plugin work with CUDA 9.0

https://www.dataiku.com/dss/plugins/info/deeplearning-image.html

I make use of preinstalled image on GCP with CUDA 9.0 installed, I dont want to setup CUDA 8.0

Regards, Longhow
0 Kudos
1 Solution
Alex_Combessie
Dataiker Alumni
Hi,

This plugin has been developed for CUDA 8.0, using tensorflow 1.4. This version of tensorflow does not support CUDA 9.0.

We have logged your feature request to update the tensorflow and CUDA versions.

If you are designing deep learning workflows in Dataiku, I would advise using directly our Deep Learning interface: https://doc.dataiku.com/dss/latest/machine-learning/deep-learning.

It is a bit more advanced (i.e. code is involved) than the plugin you mention, but it has some advantages:

- it supports both CUDA 8 and CUDA 9

- it also supports transfer learning

- you can use it for any supervised problem: image classification, text sentiment analysis, time series forecasting, etc.

- any network architecture can be used within the Keras ecosystem

- it can be exported as a real-time REST API, similarly to our other ML models

Assuming you have some experience with Keras, here is a great step-by-step tutorial on how to use it for image classification, with all code included: https://www.dataiku.com/learn/guide/visual/machine-learning/deep-learning-images.html

Hope it helps,

Alex

View solution in original post

0 Kudos
3 Replies
Alex_Combessie
Dataiker Alumni
Hi,

This plugin has been developed for CUDA 8.0, using tensorflow 1.4. This version of tensorflow does not support CUDA 9.0.

We have logged your feature request to update the tensorflow and CUDA versions.

If you are designing deep learning workflows in Dataiku, I would advise using directly our Deep Learning interface: https://doc.dataiku.com/dss/latest/machine-learning/deep-learning.

It is a bit more advanced (i.e. code is involved) than the plugin you mention, but it has some advantages:

- it supports both CUDA 8 and CUDA 9

- it also supports transfer learning

- you can use it for any supervised problem: image classification, text sentiment analysis, time series forecasting, etc.

- any network architecture can be used within the Keras ecosystem

- it can be exported as a real-time REST API, similarly to our other ML models

Assuming you have some experience with Keras, here is a great step-by-step tutorial on how to use it for image classification, with all code included: https://www.dataiku.com/learn/guide/visual/machine-learning/deep-learning-images.html

Hope it helps,

Alex
0 Kudos
longhowlam
Level 3
Author
Hi

Thanks for your answer. I am using deeplearning in dataiku, on an GCP image where CUDA 9 is set up.

The "problem" is that I wanted to use the macro in the plugin to download pretrained models. But OK, I have just manually downloaded the pretrained models. for transfer learning

Cheers,
Longhow
0 Kudos
Alex_Combessie
Dataiker Alumni
Hi,

For downloading the pretrained models, you can use the CPU version of the plugin: https://update.dataiku.com/dss/4.3/plugins/deeplearning-image-cpu/0.1.2/dss-plugin-deeplearning-image-cpu-0.1.2.zip. It does not depend on CUDA. Both versions are available in our public plugin store.

Note that all the plugin code is open source, so do not hesitate to do a pull request on our github repo if you have time and want to improve our plugins: https://github.com/dataiku/dataiku-contrib/tree/master/deeplearning-image-gpu

Cheers,

Alex
0 Kudos

Labels

?
Labels (2)
A banner prompting to get Dataiku