International Business Machines Corporation

Anamitra Roy Choudhury

Yogish Sabharwal

Ashish Verma

Sampath Dechu

Vaibhav Saxena

Monu Kedia

Thomas George

Saurabh Goyal

Pratyush Kumar

Venkatesan T Chakaravarthy

Saurabh Manish Raje

Methods, systems, and computer program products for accelerating inference of transformer-based models are provided herein. A computer-implemented method includes obtaining a machine learning model comprising a plurality of transformer blocks, a task, and a natural language dataset; generating a compressed version of the machine learning model based on the task and the natural language dataset, wherein the generating comprises: obtaining at least one set of tokens, wherein each token in the set corresponds to one of the items in the natural language dataset, identifying and removing one or more redundant output activations of different ones of the plurality of transformer blocks for the at least one set of tokens, and adding one or more input activations corresponding to the one or more removed output activations into the machine learning model at subsequent ones of the plurality of the transformer blocks; and outputting the compressed version of the machine learning model to at least one user.

Accelerating inference of transformer-based models

A computer-implemented machine learning model training method and resulting machine learning model. One embodiment of the method may comprise receiving at a computer memory training data; and training on a computer processor a machine learning model on the received training data using a plurality of batch sizes to produce a trained processor. The training may include calculating a plurality of activations during a forward pass of the training and discarding at least some of the calculated plurality of activations after the forward pass of the training.

Model training with variable batch sizing and gradient checkpoint segments

One embodiment provides a method for predicting maintenance of a redox flow battery, the method including: receiving, from a plurality of sensors, data regarding characteristics of the redox flow battery; weighting, using a processor, each of the characteristics to form an estimated state parameter for the redox flow battery; and determining, using the processor, a maintenance action for the redox flow battery using the estimated state parameter. Other aspects are described and claimed.

System and method for condition monitoring of redox flow batteries using data analytics

Methods, systems, and computer program products for load profile selection in islanding using batteries are provided herein. A computer-implemented method includes generating a usage pattern model for a battery based on (i) information pertaining to consumers of the battery, (ii) information pertaining to usage of the battery, and (iii) information pertaining to network outages, wherein the network is linked to the battery; selecting a subset of the consumers of the battery to supply energy from the battery, during a network outage, based on (i) the usage pattern model, (ii) the lifespan of the battery, (iii) performance of the network, (iv) historical consumption data attributed to each of the consumers, and (v) a predetermined minimum amount of energy required to be supplied to the consumers during a network outage; and outputting instructions to activate and/or deactivate distribution components within the network based on said selecting.

Load profile selection in adaptive dynamic islanding using batteries

A computer-implemented method of load balancing including calculating an expected cost set associated with an application-specific task of an application executing on a processing resource in a cloud computing environment, and communicating the expected cost set from the processing resource to a cloud management system. Resource mapping of applications currently executing in the cloud computing environment are retrieved, the application-specific task is assigned to a specific computational resource in the cloud computing environment based on the expected cost set and the resource mapping of applications currently executing in the cloud computing environment. A task to VM (virtual machine) assignment is determined based on the assignment of the application-specific task to the specific computational resource. The task to VM assignment is transferred from the cloud management system to the application executing on the processing resource in the cloud computing environment.

Method for improving the performance of high performance computing applications on Cloud using integrated load balancing

A non-transitory computer-implemented method of load balancing includes calculating an expected cost set associated with an application-specific task of an application executing on a processing resource in a cloud computing environment, and communicating the expected cost set from the processing resource to a cloud management system. Resource mapping of applications currently executing in the cloud computing environment are retrieved, and the application-specific task is assigned to a specific computational resource in the cloud computing environment based on the expected cost set and the resource mapping of applications currently executing in the cloud computing environment. A task to VM (virtual machine) assignment is determined based on the assignment of the application-specific task to the specific computational resource. The task to VM assignment is transferred from the cloud management system to the application executing on the processing resource in the cloud computing environment.

Growing community of inventors

Vasant Kunj, India

Anamitra Roy Choudhury