Google Inc.

Microsoft Technology Licensing, LLC

Nipun Kwatra

Vivek Raghunathan

Emily Kay Moxley

Shivakumar Venkataraman

Brett A McLarnon

David G Arthur

Rohan Jain

David J Ganzhorn

Dharma Kiritkumar Shukla

Ramachandran Ramjee

Pankaj Sharma

Muthian Sivathanu

Rimma V Nehme

Kiran S Panesar

Pranav Tiwari

Atul Katiyar

Bhalakumaaran Erode Ranganathan

Vaibhav Sharma

Shreshth Singhal

Srinidhi Viswanatha

Aditya Dalmia

The disclosure herein describes scheduling execution of artificial intelligence (AI) workloads in a cloud infrastructure platform. A global scheduler receives AI workloads associated with resource ticket values. The scheduler distributes the AI workloads to nodes based on balancing resource ticket values. Local schedulers of the nodes schedule AI workloads on resources based on the resource ticket values of the AI workloads. Based on scheduling the AI workloads, coordinator services of the local schedulers execute the distributed AI workloads on the infrastructure resources of the nodes. The disclosure further describes scheduling AI workloads based on priority tiers. A scheduler receives AI workloads, and each AI workload is associated with a priority tier indicative of a preemption priority while being executed. The AI workloads are scheduled for execution on a distributed set of nodes based on the priority tiers and then execute based on the scheduling.

Scheduler for planet-scale computing system

The disclosure herein describes platform-level checkpointing for deep learning (DL) jobs. The checkpointing is performed through capturing two kinds of state data: (i) GPU state (device state), and (ii) CPU state (host state). The GPU state includes GPU data (e.g., model parameters, optimizer state, etc.) that is located in the GPU and GPU context (e.g., the default stream in GPU, various handles created by the libraries such as DNN, Blas, etc.). Only a fraction of the GPU memory is copied because the checkpointing is done in a domain-aware manner. The 'active' memory contains useful data like model parameters. To be able to capture the useful data, memory management is controlled to identify which parts of the memory are active. Also, to restore the destination GPU to the same context/state, a mechanism is used to capture such state-changing events on an original GPU and replayed on a destination GPU.

Transparent pre-emption and migration for planet-scale computer

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, and including a computer-implemented method for providing creatives. The method comprises identifying, using one or more processors, a creative for processing, the creative including a title portion and a body portion, where the body portion includes a visual uniform resource locator (URL), the visual URL comprising a visual portion and a link to a resource. The method further comprises evaluating the visual URL for inclusion in the title portion, including determining when promotion of the visual URL satisfies one or more promotion criterion, and if so, promoting the visual URL for inclusion in the title portion. The method further comprises providing the creative including the title portion with the promoted visual URL.

Promoting content

Example processes for determining a match type include: receiving a query token from a content provider; consulting a database stored in memory to determine past performance of the query token for associated content, where the past performance includes performance information indicating how well the query token performed for different match types, where a match type indicates a way that components of the query token match components of another query token in order to achieve a token match; determining, based on the performance information, differences in performance between ones of the different match types; comparing the differences in performance to a threshold; and selecting a match type for the query token based on the comparing.

Determining match type for query tokens

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, and including a computer-implemented method that includes identifying, using one or more processors, a creative for processing. The creative includes a title portion and a body portion. The body portion includes a plurality of lines of text includes a first line of text and a second line of text. The method further includes evaluating the body portion including determining when the body portion includes a complete phrase. The evaluating includes applying a test to one or more words in the body portion. The method further includes promoting, using the one or more processors, the complete phrase into the title portion. The method further includes providing the creative includes the title portion with the promoted complete phrase.

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, and including a method for providing creatives. The method comprises identifying a creative for processing, the creative including a title portion, a body portion and optionally a reference portion. The method further comprises evaluating the body portion and the reference portion including determining when either portion includes one or more candidate words or candidate reference that can be added to the title portion. The method further comprises determining when, if promoted, the candidate one or more words or candidate reference would be redundant to content already included in the title portion. The method further comprises promoting the candidate one or more words or candidate reference into the title portion when no unacceptable redundancy is determined. The method further comprises providing the creative including the title portion with the promoted candidate one or more words or candidate reference.

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, and including a computer-implemented method for providing a creative. The method comprises identifying, using one or more processors, a creative for processing, the creative including a title portion, a body portion and optionally a reference portion. The method further comprises evaluating the body portion including determining when the body portion includes one or more words that can be added to the title portion. The method further comprises promoting, using the one or more processors, the one or more words into the title portion. The method further comprises providing the creative including the title portion with the promoted one or more words.

Growing community of inventors

Bengaluru, India

Nipun Kwatra