We’ve cut three releases of the Determined Training Platform in February so far, with some important improvements and new features to make model development easier. Here’s a recap of what’s new in Determined:

• Resource pools - Deep learning model training can get expensive, so we’re introducing resource pools as a way to more efficiently allocate GPU resources. Determined now allows you to configure your cluster to have multiple resource pools and to assign tasks to a specific resource pool so that you can use different sets of resources for different tasks. Read more about how to use resource pools here.
  • For example: In the same Determined cluster, you can create one resource pool with high-powered GPU instances, a second pool with less expensive GPUs, and a third pool with cheaper, CPU-only instances. You can then train experiments using a mix of the first and second pools, and the third pool to run TensorBoards.
• More topic guides - We’ve added walkthrough guides for common use cases.
  • The command and shell topic guide helps users use Determined in a batch manner, rather than having to write code conforming to the trial APIs.
  • The debugging how-to guide helps users fix bugs in their models (although we’re always a Slack message away to help!).
• CUDA 11 support - We’ve added support for the NVIDIA A100 GPUs for on-prem, GCP & AWS users, with new PyTorch 1.7 and TensorFlow 2.4 Docker images.

Read the full release notes for 0.14.0, 0.14.1, and 0.14.2. If you’re interested in trying Determined for the first time, check out how to install Determined.

As always, get in touch with us on our community Slack for assistance!