Abstract: Embodiments described herein provide a multimodal vision-language model. The multimodal vision-language model contains a Generalist Multimodal Transformer capable of complete multiple tasks using the same set of parameters learning from pre-training. The Generalist Multimodal Transformer allows alignment between frozen, unimodal encoders, such as image encoders and large language models. The Generalist Multimodal Transformer eliminates the need for fine-tuning the image encoders and large language models.

Abstract: System and methods are described for generating recommendations from dynamically-mapped data. In one implementation, a database system receives a first request to generate a recommendation objection and a second request to retrieve additional data to include in the recommendation object. The database system retrieves the recommendation data from a first database table. The database system identifies the additional data in a second database table that is stored separately from the first database table. The database system generates the recommendation object to include the recommendation data from the first database, and maps the additional data to one or more fields of the recommendation object.

Abstract: Methods, systems, apparatuses, devices, and computer program products are described. A flow generation service may receive a natural language input that indicates instructions for automating a task according to a first process flow. Using a large language model (LLM), the flow generation service may decompose the natural language input into a set of elements (e.g., logical actions) and connectors, where the LLM may be trained on first metadata corresponding to a second process flow that is created manually by a user. In addition, using the LLM, the flow generation service may generate second metadata corresponding to each of the set of elements based on decomposing the natural language input. The flow generation service may sequence and merge the set of elements to generate the first process flow. In some examples, the flow generation service may send, for display to a user interface of a user device, the first process flow.

Abstract: Techniques are disclosed relating to deploying application containers. A computer system may receive a request to deploy an application container onto resources of a target environment such that program code of the application container is executed. The deployment of the application container may be carried out in a mode in which the computer system does not allocate resources for the application container. The computer system may receive a request to deploy a placeholder container in the target environment. The deployment of the placeholder container may be carried out in a different mode in which the computer system allocates resources for the placeholder container. The computer system may cause an allocation of additional resources to the target environment on which to execute program code of the placeholder container. In response to the additional resources being allocated, the computer system may deploy the application container onto the additional resources for execution.

Abstract: Embodiments described herein provide a framework designed to enable personalized image generation capabilities in a pretrained text-to-image generation model. The architecture comprises two replicas of the pretrained text-to-image model—a reference UNet dedicated to extracting visual features from reference images and a base UNet for the actual image generation process. The reference UNet processes reference images to collect the features before each Self-Attention (SA) layer of the reference UNet. The base UNet's SA layers are modified to “Reference Self-Attention” (RSA) layers that allow conditioning on extra features. Using the collected reference features as input, the base UNet equipped with the RSA layers estimates the noise in the input to guide the image generation towards the reference objects.

Abstract: Methods, storage systems and computer program products create a transactionally consistent backup of a database by generating, during a time period beginning with a first time and ending with a second time, a copy of the database including multiple tables. During the time period, queries that request updates to the tables are periodically generated and conveyed to a software system managing the database, responses including updates to the tables are received responsively to the periodic queries, and the received updates are stored to a journal. A set of the updates not stored in the copy are identified, in the journal, and copy of the database is updated with the updates so as to ensure that the copy includes the transactionally consistent backup of the database.

Abstract: A multi-cloud system permits installation of software by an independent software vendor (ISV) in different manners, such as within a user space of an organization tenant of the multi-cloud system, a space of the ISV within the multi-cloud system, or a space of the multi-cloud system itself. Packaging and installation functionality of the multi-cloud system performs necessary configurations so that an ISV can simply specify how its code is to be made available to users via the multi-cloud system, without itself needing to create different configurations depending on the desired manner of usage.

Abstract: Systems, devices, and techniques are disclosed for a horizontally scalable system for managing container clusters. A monolithic management system may perform a first task on a first level of a cloud computing server system. The monolithic management system may perform a second task on a second level of a cloud computing server system. The monolithic management system may invoke instances of a container manager on a level of the cloud computing server system below the second level. The instances of the container manager may update container clusters of the cloud computing server system. The instances of the container manager may be associated with container clusters and the instances of the container manager update their associated container clusters.

Abstract: Disclosed herein are system, method, and computer program product embodiments for implementing grounding of prompt templates using record snapshot. An embodiment operates by parsing a data record to generate a hierarchical tree graph of context data corresponding to a prompt template selected based on a type of a prompt request, and the data record is identified based on a data-record reference in the prompt request. The embodiment then generates a prompt from the prompt template by grounding the prompt template with a data object comprising the hierarchical tree graph of context data. The embodiment then queries a large language model with the prompt to generate an output specific to the data record.

Abstract: Gradient-free optimization of deep learning models it performed by iteratively improving the text prompt used to instruct the deep learning model to perform a given task. An initial prompt is obtained and used to generate an initial output by the model to be optimized. This output is evaluated by an evaluation model along with reference data, evaluation criteria, etc. The evaluation model's reasoning can be used to generate a new prompt that can be passed back to the deep learning model and used to generate a new model output. This process can continue iteratively until an end condition is met.

Abstract: An agent service can perform effect detection using a convex hull representation of an input. The agent service includes at least one agent large language model (LLM) trained to perform a plurality of tasks. A representation of the input is extracted from an attention layer of the at least one agent LLM. A value space representation is determined based on the extracted representation. A task is identified from the plurality of tasks to be performed by the agent based on the value space and an embedding corresponding to the task.

Abstract: One method of identifying resource intensive queries involves receiving queries from at least one database server into an ingestion pipeline within a computing device, wherein keywords are extracted from the queries at multiple intervals; applying tokenization in the ingestion pipeline to remove uninformative terms; detecting at least one query from the received set of queries that are resource intensive at a fleet level, individual server level, or user level based on the keywords and using an enrichment analysis module, wherein the at least one query detected to be resource intensive is given a drill-down analysis to capture insights from execution profiles; and providing alerts based on the at least one query detected to be resource intensive.

Abstract: Logical implication steering in a machine learning model. In some implementations, a machine learning model with a logical implication circuit installed receives a text-based version of a prompt. The text-based version of the prompt is tokenized in an input embedding layer of the machine learning model to generate a token-based version of the prompt. From the token-based version of the prompt, a token-based version of a response is generated. This generation includes iteratively predictively selecting tokens for the token-based version of the response. The iteratively predictively selecting tokens comprises applying the logical implication circuit that, when a first vector is similar to a representation on an input to a linear mapping operation in the machine learning model, steers the generating toward a second vector. The token-based version of the response is converted to a text-based version of the response and output by the machine learning model.

Abstract: One method of receiving a web page request by a user, identifying a plurality of components to be included on the web page responsive to the web page request based on an initial assessment of the user's preferences, transmitting the web page comprising the plurality of components to the user, inserting a plug-in mechanism in response to the web page to incorporate a plurality of selectable graphical user interface (GUI) elements associated with respective components of the plurality of components of the web page, dynamically determining an updated assessment of the user's preference based at least in part on the user selection and the initial assessment of the user's preferences, and iteratively updating the respective components of the plurality of components of the web page associated with the respective ones of the plurality of selectable GUI elements.

Abstract: Database systems and methods are provided for interacting with external systems using versioning schema of the external systems. One method involves identifying, within code of a component of a graphical user interface (GUI) display, a reference to an external system and an associated version designation, which may utilize a custom versioning schema of the external system that is different from a standard versioning schema at the database system. The method constructs a versioned path for the reference to the external system using the version designation and an integration file associated with the external system, retrieves data from the external system in accordance with the version designation for the reference to the external system using the versioned path, and generates, using the retrieved data according the version designation, a graphical representation of the component within the GUI display of a virtual application provided to a client application at a client device.

Abstract: Database systems and methods are provided for managing relationships between records in a database. An exemplary method involves identifying a virtual container for a first record at the database system based on one or more fields of the first record and updating the virtual container to associate the first record with one or more additional records. In response to an operation with respect to at least one of the records, the method detects a violation of a constraint associated with the virtual container based on a relationship between respective values associated with the respective records, updates a second field of the first record to include an indication of the constraint, and generates a graphical representation of the first record that includes a graphical indication of the first constraint based at least in part on the second field of the first record.

Abstract: A voice-based agent service provides one or more voice agents to respond to voice-based requests. For example, when a question is received by the voice-based agent service it can be matched to a voice-ready knowledge base. The voice-ready knowledge base has been generated offline such that it includes concise answers without language that results in artifacts when converted to voice. These answers are organized in nodes in a conversation graph. The question can be matched to a node and then subsequent nodes can be predicted by a conversation model based on the previously traversed nodes and further user input during a real-time conversation.

Abstract: Techniques are described herein for decreasing the likelihood of out-of-domain LLM responses. In operation, it is determined whether a representation of an input token, generated during an intermediate step of the LLM operations, is in-domain or out-of-domain based on a previously detected sub-space. In addition, the response being generated by the LLM is constrained when it is determined the response being generated is not within the previously detected sub-space.

Abstract: Disclosed herein are apparatus, system and method embodiments for sending an interactive customized push notifications with customized actions. An embodiment operates by determining occurrence of an event associated with an object, forming a notification and a list of actions defined by a customer using fields from the object associated with the event, and presenting the notification and the actions to a user device that's associated to the object. Further, the action is executed in response to the user's selection.

Abstract: A method of data processing is described. The method includes receiving, at a server host, a login request that includes a clear text password for an account. The method further includes retrieving, from a database, a payload stored in association with the account. The payload may indicate a hash value of a user configured password for the account, a hashing algorithm used for generation of the hash value, an initialization vector, and a key agreement public key associated with the account. The payload may be encrypted using a secret key derived from the key agreement public key and a private key provisioned to a key protection component of the server host. The method further includes transmitting a request that includes the payload and the clear text password. The method further includes receiving a response that indicates whether the clear text password corresponds to the user configured password.