Vmware, Inc.

Facebook, Inc.

Meta Platforms, Inc.

Keith Adams

Ole Agesen

Jason Owen Evans

Sahil Rihan

Jeffrey W Sheldon

Vyacheslav Vladimirovich Malyugin

Jason Edward Weston

Boris Weissman

Maxime Austruy

Sumit Chopra

Jennifer-Ann M Anderson

Andrew John Paroski

Pratap Subrahmanyam

Narsing Krishna Vijayrao

Edwin Thur Gideon Smith

Ali-Reza Adl-Tabatabai

Spencer G Ahrens

Kinshuk Govil

Guilherme De Lima Ottoni

Eli Daniel Collins

In one embodiment, a method includes receiving, from a client system, a text input comprising one or more n-grams, determining, using a deep-learning model, a vector representation of the text input based on the one or more n-grams, determining an embedding of the vector representation of the text input in a d-dimensional embedding space, identifying one or more labels based on, for each of the one or more labels, a respective similarity of an embedding of a vector representation of the label in the embedding space to the embedding of the vector representation of the text input, and sending, to the client system in response to the received text input, instructions for presenting a user interface comprising one or more of the identified labels as suggested labels.

Predicting labels using a deep-learning model

Technology is provided for partitioning a shared unified cache in a multi-processor computer system. The technology can receive a request to allocate a portion of a shared unified cache memory for storing only executable instructions, partition the cache memory into multiple partitions, and allocate one of the partitions for storing only executable instructions. The technology can further determine the size of the portion of the cache memory to be allocated for storing only executable instructions as a function of the size of the multi-processor's L1 instruction cache and the number of cores in the multi-processor.

Partitioning shared caches

In one embodiment, a method includes accessing a first set of entities, with which a user has interacted, and a second set of entities in a social-networking system. A first set of vector representations of the first set of entities are determined using a deep-learning model. A target entity is selected from the first set of entities, and the vector representation of the target entity is removed from the first set. The remaining vector representations in the first set are combined to determine a vector representation of the user. A second set of vector representations of the second set of entities are determined using the deep-learning model. Similarity scores are computed between the user and each of the target entity and the entities in the second set of entities. Vector representations of entities in the second set of entities are updated based on the similarity scores using the deep-learning model.

Identifying entities using a deep-learning model

In one embodiment, a method includes receiving text query that includes n-grams. A vector representation of each n-gram is determined using a deep-learning model. A nonlinear combination of the vector representations of the n-grams is determined, and an embedding of the text query is determined based on the nonlinear combination. The embedding of the text query corresponds to a point in an embedding space, and the embedding space includes a plurality of points corresponding to a plurality of label embeddings. Each label embedding is based on a vector representation of a respective label determined using the deep-learning model. Label embeddings are identified as being relevant to the text query by applying a search algorithm to the embedding space. Points corresponding to the identified label embeddings are within a threshold distance of the point corresponding to the embedding of the text query in the embedding space.

Autonomous selection between multiple virtualization techniques implemented in a virtualization layer of a virtualized computer system. The virtual machine monitor implements multiple virtualization support processors that each provide for the comprehensive handling of potential virtualization exceptions. A virtual machine monitor resident virtualization selection control is operable to select between use of first and second virtualization support processors dependent on identifying a predetermined pattern of temporally local privilege dependent instructions within a portion of an instruction stream as encountered in the execution of a guest operating system.

Adaptive dynamic selection and application of multiple virtualization techniques

One embodiment of the present invention is a method of interposing operations in a computational system that includes a virtualization system executable on an underlying hardware processor that natively supports one or more instructions that transition between host and guest execution modes. The method includes introducing a hooked vector into a supervisor register block of the hardware processor, wherein the hooked vector displaces a system call handler vector otherwise set by a guest computation; read and write protecting at least the hooked vector containing portion of the supervisor register block; initiating execution of a code sequence of the guest computation on the hardware processor using one of the instructions that transition between the host and guest execution modes thereof, wherein the code sequence includes a system call and wherein upon initiation of the system call, the hardware processor transfers execution to a substitute handler in accordance with the hooked vector; and responsive to execution of the substitute handler, initiating a hooked operation and transferring control to the guest system call handler.

Interposition method suitable for hardware-assisted virtual machine

Disclosed here are methods, systems, paradigms and structures for deleting shared resources from a cache in a multi-threaded system. The shared resources can be used by a plurality of requests belonging to multiple threads executing in the system. When requests, such as requests for executing script code, and work items, such as work items for deleting a shared resource, are created, a global sequence number is assigned to each of them. The sequence number indicates the order in which the requests and work items are created. A particular work item can be executed to delete the shared resource if there are no requests having a sequence number lesser than that of the particular work item executing in the system. However, if there is at least one request with a sequence number lesser than that of the particular work item executing, the work item is ignored until the request completes executing.

Cache management in a multi-threaded environment

Scripts are incrementally compiled at runtime to generate executable code. The incremental compilation generates executable code corresponding to a basic block of the script. The executable code for a basic block of script is generated for a set of types of variables of the basic block resulting from execution of the script in response to a request. The generated executable code is stored and executed for subsequent requests if these requests result in the same types of variables for the basic block. The incremental compilation of the script is performed in a lazy fashion, such that executable code is not generated for combinations of types of variables of the script code that are not obtained from requests received at runtime. The script may dynamically generate portions of a web page that is returned in response to a request for execution of the script.

Run time incremental compilation of script code

Disclosed are a method and system for implementing an array data type of a programming language using various data structures. The disclosed method includes a plurality of implementations in which the array data type may be implemented. The implementations provide an efficient way to retrieve elements from the array, especially in the order they are inserted into the array. The data structures also minimize the computing resources required to manage and access the array. The disclosed technique also selects one of the many implementations based on criteria such as access pattern or size of the array.

Method and system for implementing an array using different data structures

One embodiment of the present invention includes a method for maintaining a shadow page table in at least partial correspondence with guest page mappings of a guest computation. The method marking with a traced write indication at least those entries of the shadow page table that map physical memory locations which themselves encode the guest page mappings, the marking identifying, for a hardware facility, a subset of memory access targets for which updates are to be recorded in a guest write buffer accessible to the virtualization system. Responsive to a coherency-inducing operation of the guest computation, the method reads from the guest write buffer and introduces corresponding updates into the shadow page table.

Growing community of inventors

San Carlos, CA, United States of America

Keith Adams