[
https://issues.apache.org/jira/browse/SINGA-406?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16691249#comment-16691249
]
Ngin Yun Chuan commented on SINGA-406:
--------------------------------------
Replies:
1.
Ok noted. If required, we will test on more CPU-only machines to ensure that
the Nvidia Docker container is compatible with any machine, even without GPU. I
can run the full train-inference flow on my Mac without GPU.
2.
Currently, I've programmed Rafiki such that the base worker image at
https://github.com/nginyc/rafiki/blob/add_pos_tagging/dockerfiles/worker.Dockerfile
installs only what Rafiki needs + `numpy==1.14.5`.
Model developers have to additionally specify the ML libraries (with version
numbers) they need to PIP-installed with the `dependencies` option during model
creation. These ML libraries will be lazily installed. They will refer to the
pages
https://nginyc.github.io/rafiki/docs/0.0.7/docs/src/user/creating-models.html
and
https://nginyc.github.io/rafiki/docs/0.0.7/docs/src/python/rafiki.client.Client.html#rafiki.client.Client.create_model.
I believe the current way I programmed aligns closely with what you said. The
only difference is that `scikit-learn` is currently not preinstalled. Also,
singa might not be installable now because it is programmed to use `pip install
...` instead of `conda install ...`.
3.
Ok noted. It is like this now.
Let me know if the documentation is not clear for any specific aspects.
The latest version is at https://nginyc.github.io/rafiki/docs/0.0.7.
> [Rafiki] Add POS tagging task & add GPU support (0.0.7)
> -------------------------------------------------------
>
> Key: SINGA-406
> URL: https://issues.apache.org/jira/browse/SINGA-406
> Project: Singa
> Issue Type: New Feature
> Reporter: Ngin Yun Chuan
> Priority: Major
>
> Refer to https://github.com/nginyc/rafiki/pull/71 for details
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
