Hi,

Like the netty native library, the convention with Maven is to use the classifier tag to specify OS specific distros.
See here for an example.

You can even use the classifier to distinguish between pure-java implementations and those that come packaged with GPU specific binaries.

I may have missed something, but it seems like we will need to install the GPU native libraries manually.

If so, can be please avoid this!

As developers, we should be able to integrate TensorFlow with a single <dependency> declaration, such that any platform specificity is fully resolved. e.g.

<dependency>
  <groupId>org.tensorflow</groupId>
  <artifactId>tensorflow</artifactId>
  <version>1.1.0</version>
  <classifier>windows-gpu</classifier>
</dependency>

It's really easy to integrate TensorFlow
All build aspects, including native libraries are fully prepackaged, expanded to filesystem, and bound, accordingly at runtime.
The entire build process on a 'clean' machine (i.e. no TensorFlow libraries installed on path) is easily possible - which makes it easy to integrate with CI servers.

This is all standard stuff for Maven and all Java devs would expect it!

kind regards
Fuzz

Add Maven artifacts to central mvnrepository #6926

Description

Metadata

Metadata

Assignees

Labels

Type

Fields

Projects

Milestone

Relationships

Development

Issue actions