Release 1.2.5 - Aug 20, 2020
It has been a while! Too long, maybe. In the meantime, GPUs and tensorflow and pytorch have raised the bar when
it comes to ML related computing. Adding automatic differentiation to compute gradients has become a must-have feature
jblas doesn't have. If you're interested in that for the JVM, have a look at deeplearning4j.org.
It seems people are still using jblas, therefore here is a small update which makes jblas work under ubuntu20.04 (amd64)
and arm64. For these new versions, I switched to openblas, as performance seems to be comparable, but compilation is
much easier.
Major changes and updates:
- updated Linux amd64 to work on ubuntu20.04. Updated libgfortran from version 3 to 4. Packaged libgfortran and
libquadmath into the JAR file so you don't have to install libgfortran.so.3 anymore.
- Switched from ATLAS to openblas. Because openblas is so much easier to compile and we can just use the libraries
that are installed. Hopefully the performance / errors are the same.
- Added libraries for arm64 (Raspberry Pis for 64bit images like ubuntu server) and AWS Graviton instances.
Hello Debian Med,
jblas (http://jblas.org/) is a Debian Java team package, but I think it
is more appropriate to ask about users here.
Upstream has released a
new version, 1.2.5; Debian currently packages 1.2.4. From the upstream release notes:
Release 1.2.5 - Aug 20, 2020
It has been a while! Too long, maybe. In the meantime, GPUs and tensorflow and pytorch have raised the bar when
it comes to ML related computing. Adding automatic differentiation to compute gradients has become a must-have feature
jblas doesn't have. If you're interested in that for the JVM, have a look at deeplearning4j.org.
It seems people are still using jblas, therefore here is a small update which makes jblas work under ubuntu20.04 (amd64)
and arm64. For these new versions, I switched to openblas, as performance seems to be comparable, but compilation is
much easier.
Major changes and updates:
- updated Linux amd64 to work on ubuntu20.04. Updated libgfortran from version 3 to 4. Packaged libgfortran and
libquadmath into the JAR file so you don't have to install libgfortran.so.3 anymore.
- Switched from ATLAS to openblas. Because openblas is so much easier to compile and we can just use the libraries
that are installed. Hopefully the performance / errors are the same.
- Added libraries for arm64 (Raspberry Pis for 64bit images like ubuntu server) and AWS Graviton instances.
My first question is, do we have users? Is the software useful to
maintain in Debian? The popcon for the package is low (around 15) and
there haven't been recent bug reports. If there are users, I will
gladly prepare an update.
And if we do update in Debian, I thought readers of this list might have opinions about whether I should switch the Debian packaging from
building against lapack to openblas.
Sysop: | Keyop |
---|---|
Location: | Huddersfield, West Yorkshire, UK |
Users: | 296 |
Nodes: | 16 (2 / 14) |
Uptime: | 88:34:06 |
Calls: | 6,658 |
Files: | 12,203 |
Messages: | 5,333,956 |