D2iQ, the company formerly known as Mesosphere, announced general availability of v1.0 it’s machine learning platform Kaptain this week. Never heard of the project? Maybe you’re more familiar with KUDO then, since that’s the name the company chose to use when introducing the tool last year.
Kaptain (or KUDO) is built on D2iQ’s Kubernetes distribution Konvoy and ropes in ML project Kubeflow to deploy and scale models. It also promises to work with Spark, Horovod, and Jupyter notebooks out of the box.
ISO C++ committee agrees on additional features for next release
The committee in charge of getting C++23 out of the door had a virtual meeting early this week and used the opportunity to adopt a few more enhancements for the specification draft. Amongst other things the group agreed on some fixes to make iterator_category and ranges work better, and welcomed an adjustment for passing enums to APIs that use the underlying type.
Starting in C++23, programmers also won’t have to add empty ( ) lambda parameter lists anymore, even when using the mutable keyword. Devs interested in more details should check out the write-up done by Herb Sutter, who is the chair of the ISO C++ standards committee.
Sysdig keeps Falco contributions flowing
After proposing the step to the Falco community in January, Container security provider Sysdig has now moved the source code for its kernel module and eBPF driver, and libraries libsinsp and libscap out of its own repos and into the Falco organisation.
According to company CTO Loris Degioanni, this completes the effort of making the Sysdig-born threat detection tool “fully free and owned by the community”. Falco was officially donated to the Cloud Native Computing Foundation in 2018. Back then the now available components were still part of Sysdig’s platform product, so a bit of detangling was needed to get them into a moveable state.
Google opens sources to Model Search system
Google’s AI department has decided to share the source code for its Model Search system. The platform is supposed to help “researchers develop the best ML models, efficiently and automatically” by following a domain agnostic approach that involves training models and evaluating the results based on the dataset and problem at hand. To arrive at the best model architecture possible, Model Search relies on knowledge distillation and weight sharing techniques.
Packer gets plugin development module
The team behind HashiCorp’s image creation tool Packer wants to lower the barrier for building Packer plugins and has therefore come up with a Plugin SDK. The latter “extracts the required plugin interfaces from the Packer repository into a standalone Go module” so that devs no longer import unneeded dependencies and have an easy way of bundling everything needed for their creation to work. It is compatible with recently released Packer 1.7, which saw the tool gaining an init command and datasources.