Impatient TensorFlow fans can try a preview of the first v2.0 alpha, barely a week after version 1.13.1 of the machine learning library shipped. The new version is pitched to bring much more ease of use. The alpha preview version is available at the TensorFlow website, along with tutorials and other documents, including a script for converting TensorFlow 1.x code to 2.0. More tutorials will follow. The team has not put out a definitive date for 2.0, but in January, said it was “coming soon”.
Intel Learns Deeply on Azure
Microsoft and Intel are partnering to bring optimized deep learning to Azure. The chip giant will offer the Intel Optimized Data Science VM for Linux (Ubuntu) via the Azure marketplace. According to the post announcing the move, “Often, default builds of popular deep learning frameworks like TensorFlow are not fully optimized for training and inference on CPU.”
Hence Intel has open-sourced its own optimizations for its Xeon processors, which is built on top of the Data Science Virtual Machine, using Intel Advanced Vector Extensions 512 (Intel AVX-512) and Intel Math Kernel Library for Deep Neural Networks (Intel MKL-DNN). The companies claimed this will deliver an average 7.7 times speed up in training times.
TeamCity hits big 70 issues
JetBrains released TeamCity 2018.2.3 this week, addressing 70 “issues”, including four “security problems” in the CI/CD server. New features include the option to delay automatic investigation assignment “to reduce false positives with flaky tests” and the provision of Docker images for Windows 1809. It did not detail the security problems that were fixed.