Vmware, Inc.

Meta Platforms, Inc.

Saleem Abdulrasool

Daniel Kerry Hiltgen

Moshe Zadka

John Powell

Patrick Devine

Mustafa Jamil

Kshitij Vijay Padalkar

Timothy Stack

Ronghua Zhang

Mallik Mahalingam

Boon Seong Ang

Jong Soo Park

Abdulkadir Utku Diril

Mikhail Smelyanskiy

Nadav Rotem

The disclosed computer-implemented method may include compiling a neural network, and the compiling may include organizing an interconnected set of nodes in a series of layers, and for each node in each layer, assigning an associated activation of a plurality of activations. Each activation may output a respective tensor of a plurality of tensors. The compiling may also include allocating memory for the activations by determining a respective memory size for each activation, and based on the respective memory size for each activation, assigning a memory block in the neural network to the activation. The method may also include, after the allocating the memory for the activations, accessing the memory blocks to perform the plurality of activations and thereby execute the neural network. Various other methods, systems, and computer-readable media are also disclosed.

Static memory allocation in neural networks

In a computer-implemented method for scheduling a plan of operations in a datacenter selection of a target from a plurality of targets in the datacenter is enabled for scheduling operations on the selected target. Selection of one or more bundles is enabled, wherein the plan of operations on the selected target are based on the one or more bundles. Dependency relationships between the selected target and other targets in the plurality of targets are determined based on the selection of one or more bundles. The plan of operations on the selected target is scheduled.

Scheduling a plan of operations in a datacenter

In a computer-implemented method for a unified graphical user interface for displaying a plan of operations in a datacenter metadata is accessed from a plurality of disparate software bundles for updating targets in a datacenter. A unified visualization of a plan of operations on the targets is displayed via a unified graphical user interface based on the accessed metadata, wherein the unified graphical user interface displays the plan of operations with a common look and feel.

Unified graphical user interface for displaying a plan of operations in a datacenter

Core dump is performed over a network without relying on network device drivers. Instead of network device drivers, firmware of network devices that is typically used during boot is preserved in memory post-boot, and one or more application program interfaces of the firmware are invoked to perform the network core dump. For ease of implementation, a network bootstrap program that has standard application program interfaces for calling into the firmware of network devices may be invoked when performing core dump over the network.

Performing network core dump without drivers

Growing community of inventors

Mountain View, CA, United States of America

Saleem Abdulrasool