Machine-learning-based browser agent

Abstract

Provided is a machine-leaming-based web browser extension that enhances web interaction and task automation. In particular, a computing system includes a browser agent extension. The browser agent extension can include a software module (e.g., a set of computer-executable instructions) that can be integrated into a web browser application to extend its capabilities. The browser extension can interact with the web browser to automate functions or interactions within the web browser that would otherwise be performed by a human user. For instance, the browser agent extension can cooperatively interoperate with other computer components such as an action service and a machine-learned sequence processing model to process observation data associated with the web browser to output actions for automatically controlling and / or otherwise interacting with the web browser application.

Description

MACHINE-LEARNING-BASED BROWSER AGENTPRIORITY CLAIM

[0001] The present application is based on and claims priority- to United States Provisional Application Number 63 / 730,849 having a filing date of December 11 , 2024. The present application claims priority- to and the benefit of each of such applications and incorporates all such applications herein by reference in their entirety.FIELD

[0002] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to a machine-leaming-based browser agent that interacts with a web browser application.BACKGROUND

[0003] A web browser application can include a software application that enables users to access, retrieve, and / or view information available on the World Wide Web or other networked computer systems or data repositories. A web browser application can interpret the code on web pages (such as HTML, CSS, and JavaScript) and display the content in a human-perceivable format. Web browser applications can also support various plugins and extensions to enhance functionality and user experience.

[0004] One existing technical challenge is to efficiently automate complex user interactions with a web browser application. Traditional methods often struggle with accurately interpreting user intent and adapting to the state changes within the browser. This limitation arises in part from the inability of conventional systems to effectively process and utilize real-time data from ongoing browser sessions.

[0005] Additionally, existing browser automation systems often do not provide sufficient transparency or interaction opportunities for users, making it difficult for users to understand and control any automated actions being performed on their behalf. This lack of intuitive user interfaces and interaction mechanisms in browser automation tools can lead to errors and inefficiencies, ultimately detracting from effective completion of the task being automated.SUMMARY

[0006] A system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions. One or more computer programs can be configured to perform particular operations or actions by virtue of including instructions that, when executed by data processing apparatus, cause the apparatus to perform the actions.

[0007] One general aspect includes a computer-implemented method. The computer - implemented method includes obtaining, by a computing system which may include one or more computing devices, observation data associated with a web browser application, where the observation data was captured by a browser agent extension associated with the web browser application. The method also includes generating, by the computing system, a model input based on the observation data. The method also includes generating, by the computing system and based on processing the model input using a machine-learned sequence processing model, a model output. The method also includes outputting, by the computing system, and based on the model output, a browser action that specifies an action to be performed relative to the web browser application. The method also includes causing, by the computing system, the browser action to be performed in the web browser application. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0008] Implementations may include one or more of the following features. The computer-implemented method may include iteratively performing the method over a plurality of browser interactions. Said iteratively performing may include automatically iteratively performing the method until a stop condition is met. The stop condition may include a user intervention. At least one model input in the iterative performance may be based on the clarifying user input. The observation data may include one or more images that depict a user interface of the web browser application. The observation data further may include textual summanes of one or more older images that depict older states of the user interface of the web browser application. The observation data may include document object model (DOM) data extracted from a web document being browsed by the web browser application. Obtaining the observation data further may include annotating the observation data with one or more identifiers that indicate interactive elements included in a web document being browsed by the web browser application. The method further may includereceiving input data associated with a user, and where generating the model input may include generating the model input based on both the observation data and the input data. The input data may include speech data or textual data input by the user. The input data may specify a requested task, and where the browser action advances the web browser application toward completion of the requested task. The browser action may be contained within a predefined action space may include a plurality of available actions that are able to be performed in the web browser application. The browser action space may include one or more of the following actions: clicking on a specified element or location; typing text into a specified input field; scrolling within a web document or a specific element of the web document navigating to a previous or forward page within a browser history of the web browser application executing a search query on a search engine; waiting for a specified duration; navigating to a specified URL; answering prompts or dialogues within the web document; hovering over a specified element; selecting an option from a dropdow n menu; pressing a specified combination of keys; opening a new browser tab; switching focus to a specific tab; closing a current browser tab; and right-clicking on a specified element or location. The browser action may include a clarify action that requests a user to provide clarifying user input. The sequence processing model is configured to perform tool use in which the sequence processing model makes calls to one or more external tools to obtain additional contextual information from the one or more external tools. Outputting, by the computing system, and based on the model output, the browser action may include: parsing the model output into a thought output and an action output; and generating the browser action based on the action output. The computer-implemented method may include causing, by the computing system, a thought output portion of the model output to be provided as an output to a user. The browser agent extension may include actuation logic that performs the browser action in the w eb browser application. The method further may include maintaining, by the computing system, current session data associated with a current browsing session, and generating, by the computing system, the model input may include generating, by the game companion agent, the model input based at least in part on the current session data. The method further may include maintaining, by the computing system, prior session data associated with one or more prior browsing sessions, and generating, by the computing system, the model input may include generating, by the computing system, the model input based at least in part on the prior session data. The sequence processing model may have been conditioned on one or more workflow demonstrations that demonstrate sequences of interactions with the web browser application. The sequence processing model may have been trained on one or moreworkflow demonstrations that demonstrate sequences of interactions with the web browser application. The method further may include: causing, by the computing system, a dedicated user interface window, panel, or element to be rendered within a user interface of the web browser application, where the dedicated user interface window, panel, or element depicts interactions, activities, or other aspects of the browser agent extension. The computer- implemented method may include: injecting HTML code into the web browser application to cause the web browser application to visualize one or more visual elements within a user interface of the web browser application. The visual elements depict performance of the browser action. The computer-implemented method may include: visualizing an agent cursor associated with the browser agent extension. The computer-implemented method may include: visualizing a heads up display of keystrokes associated with the browser agent extension. Causing, by the computing system, the thought output portion of the model output to be provided as the output may include causing, by the computing system, an audio output that verbalizes the thought output portion of the model output. The stop condition may include a task completion. The observation data can include a sequence of screenshots or pixel data capturing a visual state of the web browser application. Generating the model input can include processing the screenshots or pixel data using a multimodal vision model. The browser agent extension can interact with the web browser application via an application programming interface to capture the observation data and inject the browser action. The web browser application and the browser agent extension can be executed on a remote server computing system distinct from a client device associated with a user. Causing the browser action to be performed can include executing the browser action within the web browser application on the remote server computing sy stem. The method can further include generating, by the remote server computing system, a visual stream of the user interface of the web browser application; and transmitting the visual stream to the client device for display to the user. The browser agent extension can operate on a client device associated with a user to capture the observation data. The web browser application can be a headless browser instance executing without rendering a graphical user interface on a display device. The browser agent extension can monitor a programmatic state of the headless browser instance to generate the observation data. Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.

[0009] One general aspect includes a computing system. The computing system includes a browser agent extension configured to collect observations from a web browserapplication. The system also includes an action service. The system also includes where the browser agent extension is configured to obtain observation data associated with the web browser application and to provide the observation data to an action service. The system also includes where the action service is configured to obtain the observation data, generate a model input based on the observation data, process the model input with a machine-learned sequence processing model to generate a model output, generate a browser action based on the model output, and provide the browser action to the browser agent extension. The system also includes where the browser agent extension is configured to perform the browser action within the web browser application. In some implementations, the browser agent extension is integrated into the web browser application. In some implementations, the browser agent extension is executed on a server system. Other embodiments of this aspect include corresponding computer systems, apparatus, and computer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0010] One or more non-transitory computer-readable media can collectively store the browser agent extension described herein.

[0011] Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a block diagram illustrating an example computing system with a browser agent extension according to example implementations of aspects of the present disclosure;

[0013] Figures 2A-Q are graphical diagrams illustrating example user interfaces associated with a browser agent extension according to example implementations of aspects of the present disclosure;

[0014] Figure 3 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0015] Figure 4 is a block diagram of an example processing flow for using machine- learned model(s) to process input(s) to generate output(s) according to example implementations of aspects of the present disclosure;

[0016] Figure 5 is a block diagram of an example sequence processing model according to example implementations of aspects of the present disclosure;

[0017] Figure 6 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example implementations of aspects of the present disclosure;

[0018] Figure 7 is a block diagram of an example model development platform according to example implementations of aspects of the present disclosure;

[0019] Figure 8 is a block diagram of an example training workflow for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0020] Figure 9 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example implementations of aspects of the present disclosure;

[0021] Figure 10 is a block diagram of an example networked computing system according to example implementations of aspects of the present disclosure;

[0022] Figure 11 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure; and

[0023] Figure 12 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure.DETAILED DESCRIPTION

[0024] Example aspects of the present disclosure are directed to a machine-leaming- based web browser extension that enhances web interaction and task automation. In particular, one example aspect of the present disclosure is directed to a computing system that includes a browser agent extension. The browser agent extension can include a software module (e.g.. a set of computer-executable instructions) that can be integrated into a web browser application to extend its capabilities. The browser extension can interact with the web browser to automate functions or interactions within the web browser that would otherwise be performed by a human user. For instance, the browser agent extension can cooperatively interoperate with other computer components such as an action service and a machine-learned sequence processing model to process observation data associated with the web browser to output actions for automatically controlling and / or otherwise interacting with the web browser application. Thus, some example systems can include a browser agent extension that can capture and interpret user actions and the state of the browser. For example, the browser agent extension can take screenshots, list interactable elements, andreceive and execute complex commands from an action sen-ice (e.g., which may be executed at a backend server).

[0025] In particular, a computing system can obtain observation data from a web browser application. For example, this data can be captured by the browser agent extension. The observation data can include, for example, screenshots or Document Object Model (DOM) information. The computer system can then generate a model input from this observation data. The model input can be processed using a machine-learned sequence processing model to produce a model output. Based on this model output, the computing system can determine and output a specific browser action, such as clicking a link or filling a form. This action can then be executed, for example by the browser agent extension, within the web browser application, enhancing the interaction capability of the browser in response to the observed data.

[0026] In some implementations, the method is iteratively performed over multiple browser interactions. This iterative process can continue automatically until a defined stop condition is met. For example, the system may continue to execute browser actions, such as navigating through web pages or filling out forms, based on continuously-updated observation data until the user interv enes and / or a specific task is completed. This method allows for extended automation sequences that can handle complex tasks without requiring constant user input.

[0027] In some implementations, the iterative process is paused in response to a user intervention (e.g., which acts as a stop condition). Following the intervention, the system can receive clarifying user input, which may be in the form of text or voice commands specifying further details and / or modifications to the ongoing task(s). After incorporating this clarify ing input, the system can resume the iterative performance, utilizing the new input to refine or alter the subsequent actions. For example, if a user notices an error or wishes to change the direction of task execution, they can intervene, provide the necessary instructions, and the system will adjust its operations accordingly.

[0028] In some implementations, the user interface can facilitate both explicit and implicit interventions into the automated process executed by the browser extension. Explicit interventions may include user actions such as clicking a “Stop” button or other interface elements designed to halt or modify the automation tasks directly. On the other hand, implicit interventions can occur when the user performs typical browsing actions that signify a desire to resume manual control, such as changing tabs, closing the current tab, or interacting with web elements outside the scope of the automated tasks. Thus, in some implementations, thebrowser extension agent can automatically pause or be paused when a tab in which the browser extension agent is operation is otherwise taking actions is no longer in the user’s foreground of the browser application (e.g., because the user switched to a different tab). These interventions allow the system to seamlessly transition control between the automated processes and the user.

[0029] In some implementations, the iterative process of performing browser actions is automatically halted upon the completion of a predefined task. This stop condition can be determined by the successful execution of all steps involved in the task, as defined by the user or the underlying machine-learned model. For example, if the task involves filling out and submitting an online form, the process can be stopped once the form is successfully submitted.

[0030] In some implementations, the observation data can include images that depict the current user interface of the web browser application. These images can be screenshots that show the visible state of the browser at a given moment, allowing the system to analyze and understand the elements present on the screen, such as buttons, text fields, and other interactive components.

[0031] In some implementations, the observation data can also include textual summaries of older images that depict previous states of the user interface of the web browser application. These summaries can provide a historical context that helps the system understand changes over time within the browser’s environment. For example, if a user frequently navigates to certain web pages or interacts with specific elements, these summaries can help the system predict likely future actions and optimize the automation process accordingly.

[0032] In some implementations, the observation data can include Document Object Model (DOM) data extracted from the web document being browsed by the web browser application. The DOM data includes structured information about the elements present in the web document, such as their types, attributes, and hierarchical relationships. This data enables the system to precisely identify and interact with various elements of the web page.

[0033] In some implementations, obtaining the observation data can further include annotating the observation data with one or more identifiers that indicate interactive elements included in a web document being browsed by the web browser application. These identifiers can be tags or labels that mark elements such as links, buttons, input fields, and other components that can be interacted with. This annotated data helps the system to efficientlylocate and interact with these elements during the automation process, reducing errors and improving the overall efficiency of the task execution.

[0034] In some implementations, the computing system also receives input data associated with a user. This input data can then be utilized along with the observation data to generate the model input. As examples, this input data can include speech data or textual data provided by the user, for example specifying preferences, commands, or feedback related to the task being automated. For example, a user can input specific instructions through a text field or voice commands that direct the browser extension to perform a particular action, such as navigating to a certain webpage or filling out a form with provided details.

[0035] Thus, in some implementations, the input data provided by the user explicitly specifies a requested task, and the browser action generated by the system is designed to advance the web browser application toward the completion of this requested task. For example, if a user inputs a request to book a flight, the browser extension can automatically fill in the necessary details on a travel booking website and navigate through the booking process. This approach ensures that the actions taken by the browser extension are directly aligned with the user’s objectives, thereby streamlining the task completion process.

[0036] In some implementations, the extension and related components can be configured to generate a browser action that is defined within a pre-defined action space that includes a plurality of available actions that can be performed within the web browser application. This action space can include a number of common browser interactions, such as clicking on specified elements or locations, typing text into specified input fields, and scrolling within a web document or specific elements of the document. Additionally, actions may include navigating through the browser’s history, executing search queries, waiting for a specified duration, navigating to specific URLs, and interacting with prompts or dialogues within the web document. More complex actions can also be performed, such as hovering over elements, selecting options from dropdown menus, pressing specified combinations of keys, opening new browser tabs, switching focus between tabs, closing current browser tabs, and right-clicking on specified elements or locations. This defined set of actions allows the browser extension to automate a wide variety of tasks effectively.

[0037] In some implementations, the system can output a browser action that is a clarify action that requests a user to provide clarifying user input. This action is particularly beneficial when the system requires additional information to proceed accurately with a task or when there is ambiguity in the user’s initial instructions. For example, if a user instructs the system to book a flight but does not specify the destination, the browser extension canprompt the user to clarify by asking for the specific city or airport they wish to travel to. This clarifying action ensures that the automated processes align closely with the user's intentions and also improves the efficiency of the automated task performance.

[0038] In some implementations, the clarify action can be an action that is selectively output by the machine learning model. Additionally or alternatively, the clarify action can be triggered heuristically when certain conditions are met (e.g., when observation data contains certain content or types of content). As one example, a clarify action can be issued (e.g., either by the model and / or by some pre- or post-processing logic) when the observation data indicates that the current browser state contains any one or more of the following content types: Terms of Service (TOS), acknowledgements or indications of user consent, payment details or form of payment details, login information, personal data, and / or any other information that should be reviewed by the human user. In some implementations, the user can be given controls which enable the user to define certain categories or types of information for which the browser agent extension should pause and ask for human clarification or intervention.

[0039] In some implementations, the sequence processing model can be configured to perform tool use, where it makes calls to one or more external tools to obtain additional contextual information. This capability allows the model to enhance its understanding and decision-making processes by integrating data from various sources. For example, the model can access a weather forecasting tool to determine weather conditions for a travel booking task or consult a currency conversion tool when processing transactions in different currencies.

[0040] In some implementations, generating the browser action based on the model output can include parsing the model output into a thought output and an action output. The browser action can then be generated based on the action output. This parsing allows the system to clearly delineate between the model reasoning process (thought output) and the actionable steps (action output).

[0041] In some implementations, the method can further include providing the thought output portion of the model output as an output to the user. This feature allows users to understand the rationale behind the actions proposed or taken by the system. For instance, if the browser extension decides to fill out a form in a particular way, the thought output can explain why certain choices were made based on the input data and observed web elements, offering transparency and potentially increasing user trust in the automated system.

[0042] In some implementations, the thought output can be provided to the user as an audio output that verbalizes the thought output portion of the model output. This auditory- feedback can be particularly useful in scenarios where the user is multitasking or prefers auditory communication over visual. For example, while browsing, a user can receive spoken explanations about what the system is doing or planning to do next, such as explaining why it is navigating to a certain page or filling out specific parts of a form.

[0043] In some implementations, the browser agent extension can include actuation logic that is responsible for performing the browser action within the web browser application. This actuation logic is designed to interpret the action output from the system’s model and execute corresponding actions directly in the browser. For example, if the action output specifies clicking a button, the actuation logic can locate the button within the DOM of the webpage and trigger a click event programmatically. Similarly, if the action involves filling out a form, the logic can identify the relevant input fields and populate them with the specified data. This capability ensures that the browser extension can autonomously perform complex sequences of actions based on the processed model inputs.

[0044] In some implementations, the computing system maintains current session data associated with an ongoing browsing session. This session data can encompass a variety of information such as URLs visited, time spent on each page, interactions within the pages, and any inputs provided by the user during the session. Additionally, the computing system can use this current session data to generate the model input. For example, if the user has been comparing products on different e-commerce sites, the session data can include the products viewed, and the model input could then suggest or automate actions related to these products, such as pulling up reviews.

[0045] In some implementations, the computing system maintains prior session data associated with one or more previous browsing sessions. This prior session data can include information such as previously visited websites, user preferences, historical search queries, and past interactions within the browser. Additionally, the computing system can utilize this accumulated prior session data to generate the model input. For example, if the user has left reviews for certain content in the past, this data can inform the model to identify new items of content within its automated activities.

[0046] In some implementations, the sequence processing model can be conditioned (e.g., as few shot examples) and / or trained on one or more workflow demonstrations that illustrate sequences of interactions with the web browser application. These demonstrations can serve as practical examples or scenarios which show how various tasks can be performedwithin the browser environment, such as completing an online purchase, filling out forms, or navigating through a series of web pages. By conditioning and / or training the model on these workflow demonstrations, it can leam to recognize patterns of user behavior and the logical sequence of actions required to complete specific tasks.

[0047] In some implementations, the browser agent extension (and / or associated components) can cause the web browser application to render a dedicated user interface window, panel, or element within the user interface of the web browser application. This dedicated interface can display interactions, activities, and / or other aspects of the browser agent extension. For example, it can show a log of automated actions taken by the extension, provide options to pause or modify these actions, illustrate model analysis or chain-of- thought, and / or display real-time status updates of ongoing processes. Users can directly observe and control the workflow executed by the extension.

[0048] In some implementations, the browser agent extension (and / or associated components) can inject HTML code into the web browser application to enable the visualization of one or more visual elements within the user interface of the web browser. These visual elements can depict the performance of the browser action, providing a visual representation of the tasks being executed by the browser extension. For example, if the browser action includes filling out a form, the injected HTML code can display a progress bar or highlight the fields being populated in real time. This visual feedback can enhance user understanding of what the browser extension is doing at any given moment, contributing to a more interactive and engaging user experience. This feature allows users to visually track the extension’s activities, thereby making the process more transparent and comprehensible.

[0049] In some implementations, the visual elements can include visualizing an agent cursor associated with the browser agent extension, which provides a visual indication of the automated actions being performed by the extension within the web browser. This agent cursor can move across the screen to indicate areas of interaction, such as clicking or selecting text, effectively mirroring human cursor movements but driven by the automated processes of the extension.

[0050] In some implementations, the agent cursor described above may have one or more visual and / or behavioral characteristics that differ from the system cursor (e g., the humanly-controlled cursor). For example, the agent cursor can have a different color, shape, size, visual pattern, highlighting, movement pattern, speed, and / or other characteristic(s) relative to the human cursor. The use of alternative characteristic(s) for the agent cursor improves the ability of the user to discern or distinguish actions demonstrated as beingperformed by the agent from actions performed by the user. In some implementations, the agent cursor may be presented in place of the system cursor. For example, the system cursor may be visually removed from the interface while the agent cursor is visualized. Additionally or alternatively, the agent cursor and the system cursor may be shown concurrently to each other (e.g., with differing characteristic(s)).

[0051] As another example, the visual elements can include visualizing a heads-up display (HUD) of keystrokes associated with the browser agent extension. This HUD can show keystrokes or command sequences being executed by the extension, such as typing in a search field or navigating through menus using keyboard shortcuts.

[0052] The example visualizations described above enhance the transparency of the automated actions, allowing users to see in real-time how the browser extension interacts with web pages, which can be particularly useful for debugging purposes or for enhancing user trust and understanding of the automation process.

[0053] Thus, in some implementations, a computing system can include a browser agent extension housed within a web browser application, alongside an action service. The browser agent extension is configured to collect observation data from the web browser application, such as user interactions, page content, and browser state, and then transmit this data to the action service. The action service is designed to receive this observation data, utilize it to generate a model input, and process this input with a machine-learned sequence processing model to produce a model output. From this output, a specific browser action is generated and relayed back to the browser agent extension. The browser agent extension then executes this browser action within the web browser application, effectively automating tasks based on the processed data. This setup allows for a streamlined interaction between data collection, processing, and action implementation, enhancing the automation capabilities of the browser extension.

[0054] In some implementations, the architecture of the system can vary to accommodate different operational needs and / or infrastructure capabilities. For example, the action service can be implemented by one server system, which is responsible for receiving observation data from the browser agent extension and generating appropriate browser actions. Concurrently, the machine-learned sequence processing model that processes the input data to generate the model output can be hosted on a separate second server system, specialized for handling computationally intensive tasks. This separation can optimize the performance and scalability of each component by allowing them to operate on hardware best suited to their processing requirements.

[0055] Furthermore, in some implementations, the browser agent extension itself can be implemented within the code of the web browser application on the user's device. This setup allows for direct interaction with the web browser environment, facilitating real-time data capture and action implementation within the user’s browser session. The division of components across different systems and devices can enhance system efficiency by distributing the workload in a manner that leverages the strengths of each component.

[0056] Other configurations are also possible. For example, all components — the action service, the machine-learned model, and the browser agent extension — can be integrated into a single server system for environments where centralized control and maintenance are preferred. Alternatively, the model could be implemented in a cloud-based environment to take advantage of scalable computing resources, while the action service and the browser agent extension remain on-device. Other configurations are possible as well.

[0057] Specifically, as one example, the browser agent extension and / or the web browser application can operate entirely on a server computing system (e.g., a cloud-based server or a virtual desktop infrastructure), rather than on a local client device. In some implementations of this “remote browsing” or “server-side” configuration, the computing system hosting the browser executes the rendering engine and the agent logic remotely. The user’s local device functions as a “thin client” or display terminal, receiving a visual stream (e.g., a video feed, a pixel stream, or a remoted DOM tree) of the browser interface. User inputs (e.g., mouse clicks, keystrokes, and voice commands) are captured at the local device and transmitted over a network to the server system. The server system then injects these inputs into the server-side browser application or the browser agent extension for processing.

[0058] In some of such server-side configurations, the “browser agent extension"’ need not be structured as a traditional client-side plug-in (e.g., a .crx or .xpi file). Instead, the browser agent extension may comprise server-side code, scripts, or containers that interface directly with the browser instance running on the server. For example, the agent may interact with the browser via a direct API connection, or via internal hooks within a headless browser environment. In this context, the term “extension” broadly encompasses any software module or logic integrated with or operating alongside the browser application to facilitate the automated capture of observation data and the execution of browser actions, regardless of whether that logic is installed via a consumer-facing extension store or deployed as part of a backend infrastructure.

[0059] Additionally, implementing the browser agent extension on a server system enables the utilization of “headless” browsers for task automation. In a headlessconfiguration, the web browser application may not render a GUI visible to a user in realtime. Instead, the browser agent extension monitors the state of the headless browser programmatically (e.g., by analyzing the DOM or accessibility tree directly) and executes browser actions to complete tasks in the background. If user intervention or clarification is required, the system may dynamically instantiate a visible session or generate a visual snapshot (e.g., a screenshot or a rendered frame) to send to the user’s client device, requesting the necessary input before resuming the headless operation. This approach allows for highly scalable, parallelized task execution without consuming the computational resources of the user’s local device.

[0060] Furthermore, separating the browser agent execution from the client device provides enhanced security and data control. Because the browser actions and the observation data (which may include sensitive website content) are processed within the secure environment of the server system, the sensitive data need not be transferred to the client device. The browser agent extension can be configured to filter, redact, or sanitize the visual stream sent to the user, ensuring that only necessary information is displayed. This architecture is particularly beneficial for enterprise environments where the browser agent extension acts as a managed policy enforcer or an automated assistant that operates within a secure corporate firewall, while the user accesses the results from an unmanaged or low- power external device.

[0061] Moreover, in embodiments utilizing a computer-using agent or vision-based model, the browser agent extension may be defined by its function as the capture and actuation interface, regardless of its user-facing status. For example, in a cloud-based or sandboxed environment where the agent operates on a virtual browser, the browser agent extension can be or include the internal tooling, scripts, or protocol hooks configured to capture visual state data. In this context, the observation data can be or include continuous screenshots or pixel streams of the brow ser window- rather than, or in addition to, HTML code. The brow ser agent extension can capture these visual observations and transmit them to the multimodal model.

[0062] Conversely, in some implementations, when the model determines an action based on visual processing (e.g., identifying a button’s location visually), the browser agent extension can receive the corresponding coordinate-based instructions (e.g., X / Y coordinates) and programmatically inject the mouse clicks or keystrokes into the browser at those specific locations. Thus, the extension can include the component responsible for this screenshot-to- coordinate interface loop.

[0063] Another aspect of the present disclosure can improve the efficiency of models processing textual data in user interface (UI) action and screen understanding applications. In particular, standard tokenizers utilized in these models (e.g., which may include large language models, sequence processing models, etc.) may encounter challenges with HTML / DOM-heavy tasks due to suboptimal compression, which can lead to increased latency. This inefficiency primarily arises because HTML tags are infrequently encountered during the training of these tokenizers, resulting in their inadequate recognition as standard tokens.

[0064] To address this issue, some example implementations can incorporate HTML tags as user-defined tokens within the tokenizer’s vocabulary. For example, HTML tags such as . <html>, <body>, , , <bbox>, , , . <iframe>, <footer>, along with their respective closing tags, can be added as additional tokens. By integrating these tags directly into the tokenizer, the model can more readily recognize and process HTML / DOM-heavy content. This increased ability can reduce latency and improve the performance of Ul-action and screen understanding tasks.

[0065] The systems and methods of the present disclosure provide a number of technical effects and benefits. As one example, the proposed systems provide improved browser interaction and automation. The integration of a browser agent extension and an action service within the computing system allows for dynamic interaction with a web browser application, the system can automatically generate and execute browser actions based on real-time user interactions and browser states. The ability to automatically perform such actions directly addresses technical challenges related to automating complex sequences of web interactions.

[0066] As another example technical effect, the proposed technology introduces advanced user interface elements, such as dedicated windows, panels, or elements that visualize the operations of the browser agent extension. These visualizations can additionally or alternatively include agent cursors and heads-up displays for keystrokes. The visualizations and other user interface elements not only provide users with a clear view of the automated actions but also facilitate better user control and understanding of the automation process. By enhancing the user interface in this manner, the technology addresses technical issues related to user interaction with automated systems. Increased transparency and controllability in the system can lead to more targeted and efficient use of computational resources by allowing users to closely monitor and adjust the actions taken by the browser extension. For example, by visualizing the actions and enabling modifications, users canprevent unnecessary tasks from executing, thereby reducing redundant computational efforts and enhancing overall system performance.

[0067] As another example technical effect and benefit, integrating this technology directly as a browser extension can reduce computational consumption by leveraging the native functionalities and resources of the web browser, which eliminates the need for duplicative processing and data transfer that a standalone system might require. For example, by utilizing the browser’s inherent capabilities to manage DOM interactions and network requests, the extension can perform tasks more efficiently than a separate system that must independently replicate these functions.

[0068] Figure 1 illustrates a schematic diagram of a computing system configured to enhance web interaction and task automation through a browser agent extension and an action service. A web browser application 102 encompasses a browser engine 104, which interacts with various components to facilitate web browsing and data processing. The browser agent extension 106, integrated within the web browser application 102, captures observation data 108, which can include DOM data, screengrabs, and / or other similar information. The observation data 108 can, in some cases, also be supplemented with user input 109.

[0069] The observation data 108 is sent to the action service 110, which processes this data to generate a model input 112. The model input 112 is then processed by a machine- learned model 114, resulting in a model output 116 that is used to select one or more browser actions 120. These actions are executed within the web browser application 102 through the actuation logic 122, which performs the actuations 124 affecting the browser output 126. Additionally, the system can access external tools 118 to enhance the processing capabilities of the machine-learned model 114. The entire process can be supported by a memory’ layer 1 19, which stores data and outputs throughout the interaction cycle(s).

[0070] In some implementations, the system illustrated in Figure 1 can operate iteratively to perform multiple browser interactions. This iterative process can continue automatically until a defined stop condition is met, such as user intervention or the completion of a specific task. For example, the system may continue to execute browser actions, such as navigating through web pages or filling out forms, based on continuously- updated observation data 108 until the user interv enes and / or a specific task is completed. This method allows for extended automation sequences that can handle complex tasks without requiring constant user input.

[0071] If user intervention occurs, which acts as a stop condition, the iterative process is paused. Following the intervention, the system can receive clarifying user input 109. which may be in the form of text or voice commands specifying further details and / or modifications to the ongoing task(s). After incorporating this clarifying input, the system can resume the iterative performance, utilizing the new input to refine or alter the subsequent actions. For example, if a user notices an error or wishes to change the direction of task execution, they can intervene, provide the necessary instructions, and the system will adjust its operations accordingly.

[0072] The user interface can facilitate both explicit and implicit interventions into the automated process executed by the browser extension 106. Explicit interventions may include user actions such as clicking a “Stop” button or other interface elements designed to halt or modify the automation tasks directly. On the other hand, implicit interventions can occur when the user performs typical browsing actions that signify a desire to resume manual control, such as changing tabs, closing the current tab, or interacting with w eb elements outside the scope of the automated tasks. These interventions allow the system to seamlessly transition control between the automated processes and the user.

[0073] The iterative process of performing browser actions 120 can be automatically halted upon the completion of a predefined task. This stop condition can be determined by the successful execution of all steps involved in the task, as defined by the user or the underlying machine-learned model 114. For example, if the task involves filling out and submitting an online form, the process can be stopped once the form is successfully submitted.

[0074] The observation data 108 can include images that depict the current user interface of the web brow ser application 102. These images can be screenshots that show- the visible state of the browser at a given moment, allowing the system to analyze and understand the elements present on the screen, such as buttons, text fields, and other interactive components. Additionally, the observation data 108 can also include textual summaries of older images that depict previous states of the user interface of the web browser application 102. These summaries, for example which may be stored in memory layer 119, can provide a historical context that helps the system understand changes over time w ithin the browser’s environment. For example, if a user frequently navigates to certain w eb pages or interacts with specific elements, these summaries can help the system predict likely future actions and optimize the automation process accordingly.

[0075] The observation data 108 can include Document Object Model (DOM) data extracted from the web document being browsed by the web browser application 102. TheDOM data includes structured information about the elements present in the web document, such as their types, attributes, and hierarchical relationships. This data enables the system to precisely identify and interact with various elements of the web page. Obtaining the observation data can further include annotating the observation data with one or more identifiers that indicate interactive elements included in a web document being browsed by the web browser application 102. These identifiers can be tags or labels that mark elements such as links, buttons, input fields, and other components that can be interacted with. This annotated data helps the system to efficiently locate and interact with these elements during the automation process, reducing errors and improving the overall efficiency of the task execution.

[0076] The computing system also receives input data 109 associated with a user. This input data can then be utilized along with the observation data 108 to generate the model input 112. As examples, this input data can include speech data or textual data provided by the user, for example specifying preferences, commands, or feedback related to the task being automated. For example, a user can input specific instructions through a text field or voice commands that direct the browser extension 106 to perform a particular action, such as navigating to a certain webpage or filling out a form with provided details.

[0077] In some implementations, the input data provided by the user explicitly specifies a requested task, and the browser action 120 generated by the system is designed to advance the web browser application 102 toward the completion of this requested task. For example, if a user inputs a request to book a flight, the browser extension 106 can automatically fill in the necessary details on a travel booking website and navigate through the booking process. This approach ensures that the actions taken by the browser extension 106 are directly aligned with the user’s objectives, thereby streamlining the task completion process.

[0078] In some implementations, the extension and related components can be configured to generate a browser action 120 that is defined within a pre-defined action space that includes a plurality of available actions that can be performed within the web browser application 102. This action space can include a number of common browser interactions, such as clicking on specified elements or locations, typing text into specified input fields, and scrolling within a w eb document or specific elements of the document. Additionally, actions may include navigating through the browser’s history, executing search queries, waiting for a specified duration, navigating to specific URLs, and interacting with prompts or dialogues within the web document. More complex actions can also be performed, such as hoveringover elements, selecting options from dropdown menus, pressing specified combinations of keys, opening new browser tabs, switching focus between tabs, closing current browser tabs, and right-clicking on specified elements or locations. This defined set of actions allows the browser extension 106 to automate a wide variety of tasks effectively.

[0079] In some implementations, the system can output a browser action 120 that is a clarify action that requests a user to provide clarifying user input. This action is particularly beneficial when the system requires additional information to proceed accurately with a task or when there is ambiguity’ in the user’s initial instructions. For example, if a user instructs the system to book a flight but does not specify the destination, the browser extension 106 can prompt the user to clarify by asking for the specific city or airport they wish to travel to. This clarifying action ensures that the automated processes align closely with the user’s intentions and also improves the efficiency of the automated task performance.

[0080] In some implementations, the model 114 can be configured to perform tool use, where it makes calls to one or more external tools 118 to obtain additional contextual information. This capability allows the model to enhance its understanding and decisionmaking processes by integrating data from various sources. For example, the model can access a weather forecasting tool to determine weather conditions for a travel booking task or consult a currency conversion tool when processing transactions in different currencies.

[0081] In some implementations, generating the browser action 120 based on the model output 116 can include parsing the model output into a thought output and an action output. The browser action can then be generated based on the action output. This parsing allows the system to clearly delineate between the model reasoning process (thought output) and the actionable steps (action output).

[0082] In some implementations, the method can further include providing the thought output portion of the model output 116 as an output to the user. This feature allows users to understand the rationale behind the actions proposed or taken by the system. For instance, if the browser extension 106 decides to fill out a form in a particular way, the thought output can explain why certain choices were made based on the input data 109 and observed web elements, offering transparency and potentially increasing user trust in the automated system.

[0083] In some implementations, the thought output can be provided to the user as an audio output of browser output 126 that verbalizes the thought output portion of the model output 116. This auditory feedback can be particularly’ useful in scenarios where the user is multitasking or prefers auditory communication over visual. For example, while browsing, auser can receive spoken explanations about what the system is doing or planning to do next, such as explaining why it is navigating to a certain page or filling out specific parts of a form.

[0084] In some implementations, the browser agent extension 106 can include actuation logic 122 that is responsible for performing the browser action 120 within the web browser application 102. This actuation logic is designed to interpret the action output from the system's model 114 and execute corresponding actions directly in the browser. For example, if the action output specifies clicking a button, the actuation logic 122 can locate the button within the DOM of the webpage and trigger a click event programmatically. Similarly, if the action involves filling out a form, the logic can identify the relevant input fields and populate them with the specified data. This capability ensures that the browser extension can autonomously perform complex sequences of actions based on the processed model inputs.

[0085] In some implementations, the computing system maintains (e.g., in memory layer 119) current session data associated with an ongoing browsing session. This session data can encompass a variety of information such as URLs visited, time spent on each page, interactions within the pages, and any inputs provided by the user during the session. Additionally, the computing system can use this current session data to generate the model input 1 12. For example, if the user has been comparing products on different e-commerce sites, the session data can include the products viewed, and the model input could then suggest or automate actions related to these products, such as pulling up reviews.

[0086] In some implementations, the computing system maintains (e.g., in memory layer 1 19) prior session data associated with one or more previous browsing sessions. This prior session data can include information such as previously visited websites, user preferences, historical search queries, and past interactions within the browser. Additionally, the computing system can utilize this accumulated prior session data to generate the model input 112. For example, if the user has left reviews for certain content in the past, this data can inform the model to identify new items of content within its automated activities.

[0087] In some implementations, the sequence processing model 114 can be conditioned (e.g., as few shot examples) and / or trained on one or more workflow demonstrations that illustrate sequences of interactions with the web browser application 102. These demonstrations can serve as practical examples or scenarios which show how various tasks can be performed within the browser environment, such as completing an online purchase, filling out forms, or navigating through a series of web pages. By conditioning and / or training the model on these workflow demonstrations, it can leam to recognizepaterns of user behavior and the logical sequence of actions required to complete specific tasks.

[0088] In some implementations, the browser agent extension 106 (and / or associated components) can cause the web browser application 102 to render a dedicated user interface window panel, or element w ithin the user interface of the w eb brow ser application 102. This dedicated interface can display interactions, activities, and / or other aspects of the browser agent extension 106. For example, it can show' a log of automated actions taken by the extension, provide options to pause or modify these actions, illustrate model analysis or chain-of-thought, and / or display real-time status updates of ongoing processes. Users can directly observe and control the workflow executed by the extension.

[0089] In some implementations, the browser agent extension 106 (and / or associated components) can inject HTML code into the web browser application 102 (e.g., for execution by the browser engine 104) to enable the visualization of one or more visual elements within the user interface of the web browser application 102. These visual elements can depict the performance of the browser action 120, providing a visual representation of the tasks being executed by the browser extension 106. For example, if the browser action includes filling out a form, the injected HTML code can display a progress bar or highlight the fields being populated in real time. This visual feedback can enhance user understanding of w hat the browser extension is doing at any given moment, contributing to a more interactive and engaging user experience. This feature allows users to visually track the extension’s activities, thereby making the process more transparent and comprehensible.

[0090] In some implementations, the visual elements can include visualizing an agent cursor associated w ith the browser agent extension 106, which provides a visual indication of the automated actions being performed by the extension within the web browser application 102. This agent cursor can move across the screen to indicate areas of interaction, such as clicking or selecting text, effectively mirroring human cursor movements but driven by the automated processes of the extension.

[0091] In some implementations, the agent cursor described above may have one or more visual and / or behavioral characteristics that differ from the system cursor (e.g., the humanly -controlled cursor). For example, the agent cursor can have a different color, shape, size, visual patern, highlighting, movement patern, speed, and / or other characteristic(s) relative to the human cursor. The use of alternative characteristic(s) for the agent cursor improves the ability of the user to discern or distinguish actions demonstrated as being performed by the agent from actions performed by the user. In some implementations, theagent cursor may be presented in place of the system cursor. For example, the system cursor may be visually removed from the interface while the agent cursor is visualized. Additionally or alternatively, the agent cursor and the system cursor may be shown concurrently to each other (e.g., with differing character! stic(s)).

[0092] As another example, the visual elements can include visualizing a heads-up display (HUD) of keystrokes associated with the browser agent extension 106. This HUD can show keystrokes or command sequences being executed by the extension, such as typing in a search field or navigating through menus using keyboard shortcuts.

[0093] The example visualizations described above enhance the transparency of the automated actions, allowing users to see in real-time how the browser extension interacts with web pages, which can be particularly useful for debugging purposes or for enhancing user trust and understanding of the automation process.

[0094] Figure 3 depicts a flowchart of a method 300 for training one or more machine-learned models according to aspects of the present disclosure. For instance, an example machine-learned model can include a sequence processing model.

[0095] One or more portion(s) of example method 300 can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of example method 300 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 300 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 3 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 3 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 300 can be performed additionally, or alternatively, by other systems.

[0096] At 302, example method 300 can include obtaining a training instance. A set of training data can include a plurality of training instances divided between multiple datasets (e.g., a training dataset, a validation dataset, or testing dataset). A training instance can be labeled or unlabeled. Although referred to in example method 300 as a “training’7instance, it is to be understood that runtime inferences can form training instances when a model istrained using an evaluation of the model’s performance on that runtime instance (e.g., online training / leaming). Example data types for the training instance and various tasks associated therewith are described throughout the present disclosure.

[0097] At 304, example method 300 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can be directly obtained from the one or more machine-learned models or can be a downstream result of a chain of processing operations that includes an output of the one or more machine- learned models.

[0098] At 306, example method 300 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function. Various determinations of loss can be used, such as mean squared error, likelihood loss, cross entropy loss, hinge loss, contrastive loss, or various other loss functions. The evaluation signal can be computed using known ground-truth labels (e.g., supervised learning), predicted or estimated labels (e.g., semi- or self-supervised learning), or without labels (e.g., unsupervised learning). The evaluation signal can be a reward (e.g., for reinforcement learning). The reward can be computed using a machine-learned reward model configured to generate rewards based on output(s) received. The reward can be computed using feedback data describing human feedback on the output(s).

[0099] At 308, example method 300 can include updating the machine-learned model using the evaluation signal. For example, values for parameters of the machine-learned model(s) can be learned, in some embodiments, using various training or learning techniques, such as, for example, backwards propagation. For example, the evaluation signal can be backpropagated from the output (or another source of the evaluation signal) through the machine-learned model(s) to update one or more parameters of the model(s) (e g., based on a gradient of the evaluation signal with respect to the parameter value(s)). For example, system(s) containing one or more machine-learned models can be trained in an end-to-end manner. Gradient descent techniques can be used to iteratively update the parameters over a number of training iterations. In some implementations, performing backwards propagation of errors can include performing truncated backpropagation through time. Example method 300 can include implementing a number of generalization techniques (e.g., w eight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0100] In some implementations, example method 300 can be implemented for training a machine-learned model from an initialized state to a fully trained state (e.g., whenthe model exhibits a desired performance profile, such as based on accuracy, precision, recall, etc.).

[0101] In some implementations, example method 300 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 300 can be implemented for pre-training a machine-learned model. Pre-training can include, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety of tasks / data types.

[0102] In some implementations, example method 300 can be implemented for fine- tuning a machine-learned model. Fine-tuning can include, for instance, smaller-scale training on higher-quality (e.g.. labeled, curated, etc.) data. Fine-tuning can affect all or a portion of the parameters of a machine-learned model. For example, various portions of the machine- learned model can be "frozen" for certain training stages. For example, parameters associated with an embedding space can be “frozen” during fine-tuning (e.g., to retain information learned from a broader domain(s) than present in the fine-tuning dataset(s)). In some implementations, example method 300 uses adapter modules. Adapters can be small trainable layers that are inserted between pre-existing layers of a pre-trained model. During the fine- tuning process, the original parameters of the pre-trained model are typically frozen, and only the parameters of the adapters are updated.

[0103] In some implementations, example method 300 can be implemented to execute parameter-efficient fine-tuning methods, such as Layerwise Optimization of Residuals (LoRA). LoRA can refine pre-trained models with minimal adjustments to the original parameters. This can be achieved by introducing trainable low-rank matrices that modify the behavior of the pre-trained w eights without directly altering them. In some implementations, during fine-tuning, only these auxiliary matrices are updated, which significantly reduces the number of parameters that are trained.

[0104] An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use.

[0105] Figure 4 is a block diagram of an example processing flow for using machine- learned model(s) 1 to process input(s) 2 to generate output(s) 3.

[0106] Machine-learned model(s) 1 can be or include one or multiple machine- learned models or model components. Example machine-learned models can include neural networks (e.g., deep neural networks). Example machine-learned models can include nonlinear models or linear models. Example machine-learned models can use other architectures in lieu of or in addition to neural networks. Example machine-learned models can includedecision tree based models, support vector machines, hidden Markov models, Bayesian networks, linear regression models, k-means clustering models, etc.

[0107] Machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of the machine-learned models described above with respect to the preceding figures. For example, machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of the models described herein, etc. Although various features, variations, and implementations described below are described with respect to machine-learned model(s) 1, it is to be understood that such features, variations, and implementations are to be understood as described with respect to any other machine-learned component described herein.

[0108] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recurrent neural networks, convolutional neural networks (CNNs), diffusion models, generative-adversarial netw orks, or other forms of neural netw orks. Example neural networks can be deep neural networks. Some example machine-learned models can leverage an attention mechanism such as self-attention. For example, some example machine-learned models can include multiheaded self-attention models.

[0109] Machine-learned model(s) 1 can include a single or multiple instances of the same model configured to operate on data from input(s) 2. Machine-learned model(s) 1 can include multiple different models or multiple different model portions configured to operate on data from input(s) 2.

[0110] Machine-learned model(s) 1 can include an ensemble of different models that can cooperatively interact to process data from input(s) 2. For example, a model ensemble can include multiple models that have different attributes (e.g., different architectures, trained with different recipes, etc.). The ensemble can output an overall output based on the individual outputs of the constituent models. In this manner, for instance, the diverse constituent models can work together to provide system-level robustness by effectively aggregating over individual strengths and weaknesses of any given model. The respective individual outputs can be combined in a weighted combination, using a voting or routing mechanism, or a learned output layer (e.g., one or more feedforward or fully-connected layers).[Oi l 1] Machine-learned model(s) 1 can employ a mixture-of-experts structure. See, e.g., Zhou et a ., Mixture-of-Experts with Expert Choice Routing, ARXIV:2202.09368V2 (Oct. 14, 2022). For example, different portions of a model can leam (explicitly or implicitly)different expertise areas, with pathways through the model being selected by a learned routing mechanism that engages the appropriate expert for a given input (e.g., a given portion of an input, such as on a per-token basis). For example, a feedforward network can be sparsely activated for a given portion of an input based on an output of a routing mechanism that processes the portion of the input. In this manner, for instance, the group of activated weights can form an “expert” that is selected by the router. On each forward pass, only a subset of the total model weights may be engaged, thereby decreasing a quantity of operations performed for processing a given input compared to a densely activated model. In this manner, for instance, the expressive and interpretive power of a high-parameter-count model can be achieved with more compute-efficient forward passes.

[0112] Input(s) 2 can generally include or otherwise represent various types of data. Input(s) 2 can include one type or many different types of data. Output(s) 3 can be data of the same type(s) or of different types of data as compared to input(s) 2. Output(s) 3 can include one ty pe or many different types of data.

[0113] Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g., binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer’s central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e.g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like. Data can be raw or processed and can be in any format or schema.

[0114] In multimodal inputs 2 or outputs 3, example combinations of data types include image data and audio data, image data and natural language data, natural language data and software code data, image data and biometric data, sensor data and medical data, etc. It is to be understood that any combination of data types in an input 2 or an output 3 can be present.

[0115] An example input 2 can include one or multiple data types, such as the example data types noted above. An example output 3 can include one or multiple data types, such as the example data types noted above. The data type(s) of input 2 can be the same as ordifferent from the data type(s) of output 3. It is to be understood that the example data types noted above are provided for illustrative purposes only. Data types contemplated within the scope of the present disclosure are not limited to those examples noted above.

[0116] Figure 5 is a block diagram of an example implementation of an example machine-learned model configured to process sequences of information. For instance, an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4. An example system can pass input(s) 2 to sequence processing model(s) 4. Sequence processing model(s) 4 can include one or more machine- learned components. Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5. Input sequence 5 can include one or more input elements 5-1, 5- 2, . . . , 5-AL, etc. obtained from input(s) 2. Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7. Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7-N, etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7.

[0117] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models in the text domain are referred to as “Large Language Models,” or LLMs. See, e.g., PaLM 2 Technical Report, GOOGLE, https: / / ai.google / static / documents / palm2techreport.pdf (n d ). Other example sequence processing models can operate in other domains, such as image domains, see, e.g., Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXIV:2010. 11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et al., MusicLM: Generating Music From Text, ARXIV:2301.11325V1 (Jan. 26, 2023), biochemical domains, see. e.g., Jumper et al.. Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example. Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), relatively small models (e.g., fewer parameters, computationally lightweight, etc ), or both.

[0118] In general, sequence processing model (s) 4 can obtain input sequence 5 using data from input(s) 2. For instance, input sequence 5 can include a representation of data from input(s) 2 in a format understood by sequence processing model(s) 4. One or more machine- learned components of sequence processing model(s) 4 can ingest the data from input(s) 2, parse the data into pieces compatible with the processing architectures of sequenceprocessing model(s) 4 (e.g., via “tokenization”), and project the pieces into an input space associated with prediction layer(s) 6 (e.g., via “embedding”).

[0119] Sequence processing model(s) 4 can ingest the data from input(s) 2 and parse the data into a sequence of elements to obtain input sequence 5. For example, a portion of input data from input(s) 2 can be broken down into pieces that collectively represent the content of the portion of the input data. The pieces can provide the elements of the sequence.

[0120] Elements 5-1, 5-2, . . . . 5-M can represent, in some cases, building blocks for capturing or expressing meaningful information in a particular data domain. For instance, the elements can describe “atomic units” across one or more domains. For example, for textual input source(s), the elements can correspond to groups of one or more words or sub-word components, such as sets of one or more characters.

[0121] For example, elements 5-1, 5-2, . . . , 5-M can represent tokens obtained using atokenizer. For instance, a tokenizer can process a given portion of an input source and output a series of tokens (e.g., corresponding to input elements 5-1, 5-2, . . . , 5 -AT) that represent the portion of the input source. Various approaches to tokenization can be used. For instance, textual input source(s) can be tokenized using a byte-pair encoding (BPE) technique. See, e.g., Kudo et al., SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing, PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66-71 (October 31-November 4. 2018), https: / / aclanthology.org / Dl 8-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image.

[0122] In general, arbitrary' data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-M depicted in Figure 5 can be the tokens or can be the embedded representations thereof.

[0123] Prediction layer(s) 6 can predict one or more output elements 7-1, 7-2, . . . , 7- N based on the input elements. Prediction layer(s) 6 can include one or more machine-learned model architectures, such as one or more layers of learned parameters that manipulate and transform the input(s) to extract higher-order meaning from, and relationships between, input element(s) 5-1, 5-2, . . . , 5-M. In this manner, for instance, example prediction layer(s) 6 can predict new output element(s) in view of the context provided by input sequence 5.

[0124] Prediction layer(s) 6 can evaluate associations between portions of input sequence 5 and a particular output element. These associations can inform a prediction of the likelihood that a particular output follows the input context. For example, consider the textualsnippet, “The carpenter’s toolbox was small and heavy. It was full of .” Example prediction layer(s) 6 can identify that “It” refers back to “toolbox” by determining a relationship between the respective embeddings. Example prediction layer(s) 6 can also link “It” to the attributes of the toolbox, such as “small” and “heavy.” Based on these associations, prediction layer(s) 6 can, for instance, assign a higher probability to the word “nails” than to the word “sawdust.”

[0125] A transformer is an example architecture that can be used in prediction layer(s) 4. See, e.g., Vaswani et al.. Attention Is All You Need, ARXIV: 1706.03762V7 (Aug. 2, 2023). A transformer is an example of a machine-learned model architecture that uses an attention mechanism to compute associations between items within a context window'. The context window can include a sequence that contains input sequence 5 and potentially one or more output element(s) 7-1, 7-2, . . . , 7-N. A transformer block can include one or more attention layer(s) and one or more post-attention layer(s) (e.g., feedforward layer(s), such as a multi-layer perceptron).

[0126] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as w'ell as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information.

[0127] Output sequence 7 can include or otherwise represent the same or different data types as input sequence 5. For instance, input sequence 5 can represent textual data, and output sequence 7 can represent textual data. Input sequence 5 can represent image, audio, or audiovisual data, and output sequence 7 can represent textual data (e.g., describing the image, audio, or audiovisual data). It is to be understood that prediction layer(s) 6, and any other interstitial model components of sequence processing model(s) 4, can be configured to receive a variety of data types in input sequence(s) 5 and output a variety of data ty pes in output sequence(s) 7.

[0128] Output sequence 7 can have various relationships to input sequence 5. Output sequence 7 can be a continuation of input sequence 5. Output sequence 7 can be complementary to input sequence 5. Output sequence 7 can translate, transform, augment, or otherwise modify' input sequence 5. Output sequence 7 can answer, evaluate, confirm, or otherwise respond to input sequence 5. Output sequence 7 can implement (or describe instructions for implementing) an instruction provided via input sequence 5.

[0129] Output sequence 7 can be generated autoregressively. For instance, for some applications, an output of one or more prediction layer(s) 6 can be passed through one or more output layers (e.g., softmax layer) to obtain a probability distribution over an output vocabulary (e.g., a textual or symbolic vocabulary) conditioned on a set of input elements in a context window. In this manner, for instance, output sequence 7 can be autoregressively generated by sampling a likely next output element, adding that element to the context window, and re-generating the probability distribution based on the updated context window, and sampling a likely next output element, and so forth.

[0130] Output sequence 7 can also be generated non-autoregressively. For instance, multiple output elements of output sequence 7 can be predicted together without explicit sequential conditioning on each other. See, e.g., Saharia et al., Non-Autoregressive Machine Translation with Latent Alignments, ARXIV:2004.07437V3 (Nov. 16, 2020).

[0131] Output sequence 7 can include one or multiple portions or elements. In an example content generation configuration, output sequence 7 can include multiple elements corresponding to multiple portions of a generated output sequence (e.g., a textual sentence, values of a discretized waveform, computer code, etc.). In an example classification configuration, output sequence 7 can include a single element associated with a classification output. For instance, an output “vocabulary” can include a set of classes into which an input sequence is to be classified. For instance, a vision transformer block can pass latent state information to a multilayer perceptron that outputs a likely class value associated with an input image.

[0132] Figure 6 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task). Input sequence 8 can include various data elements from different data modalities. For instance, an input modality 10-1 can include one modality of data. A data-to-sequence model 11-1 can process data from input modality 10-1 to project the data into a format compatible with input sequence 8 (e.g., one or more vectors dimensioned according to the dimensions of input sequence 8) to obtain elements 8-1, 8-2, 8-3. Another input modality 10-2 can include a different modality of data. A data-to-sequence model 11-2 can project data from input modality 10-2 into a format compatible with input sequence 8 to obtain elements 8-4, 8-5, 8- 6. Another input modality 10-3 can include yet another different modality of data. A data-to-sequence model 11-3 can project data from input modality 10-3 into a format compatible with input sequence 8 to obtain elements 8-7. 8-8, 8-9.

[0133] Input sequence 8 can be the same as or different from input sequence 5. Input sequence 8 can be a multimodal input sequence that contains elements that represent data from different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions. Input sequence 8 can be configured to contain a plurality of elements that have P dimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween.

[0134] For example, elements 8-0. . . . , 8-9 can indicate particular locations within a multidimensional embedding space. Some elements can map to a set of discrete locations in the embedding space. For instance, elements that correspond to discrete members of a predetermined vocabulary of tokens can map to discrete locations in the embedding space that are associated with those tokens. Other elements can be continuously distributed across the embedding space. For instance, some datatypes can be broken down into continuously defined portions (e.g., image patches) that can be described using continuously distributed locations within the embedding space.

[0135] In some implementations, the expressive power of the embedding space may not be limited to meanings associated with any particular set of tokens or other building blocks. For example, a continuous embedding space can encode a spectrum of high-order information. An individual piece of information (e.g., a token) can map to a particular point in that space: for instance, a token for the word ‘‘dog’' can be projected to an embedded value that points to a particular location in the embedding space associated with canine-related information. Similarly, an image patch of an image of a dog on grass can also be projected into the embedding space. In some implementations, the projection of the image of the dog can be similar to the projection of the word “dog” while also having similarity to a projection of the word “grass.” while potentially being different from both. In some implementations, the projection of the image patch may not exactly align with any single projection of a single word. In some implementations, the projection of the image patch can align with a combination of the projections of the words “dog” and “grass.” In this manner, for instance, a high-order embedding space can encode information that can be independent of data modalities in which the information is expressed.

[0136] Task indicator 9 can include a model or model component configured to identify a task being performed and inject, into input sequence 8, an input value represented by element 8-0 that signals which task is being performed. For instance, the input value can be provided as a data type associated with an input modality and projected along with that input modality (e.g., the input value can be a textual task label that is embedded along with other textual data in the input; the input value can be a pixel-based representation of a task that is embedded along with other image data in the input; etc.). The input value can be provided as a data type that differs from or is at least independent from other input(s). For instance, the input value represented by element 8-0 can be learned within a continuous embedding space.

[0137] Input modalities 10-1, 10-2, and 10-3 can be associated with various different data types (e.g., as described above with respect to input(s) 2 and output(s) 3).

[0138] Data-to-sequence models 11-1, 11-2, and 11-3 can be the same or different from each other. Data-to-sequence models 11-1, 11-2, and 11-3 can be adapted to each respective input modality 10-1, 10-2, and 10-3. For example, a textual data-to-sequence model can subdivide a portion of input text and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-1, 8-2, 8-3, etc.). An image data-to-sequence model can subdivide an input image and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-4, 8-5, 8-6, etc.). An arbitrary data type data-to-sequence model can subdivide an input of that arbitrary data type and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7, 8-8, 8-9, etc.).

[0139] Data-to-sequence models 11-1, 11-2, and 11-3 can form part of machine- learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be jointly trained with or trained independently from machine-learned sequence processing model(s) 4. Data-to-sequence models 11-1, 11-2, and 11-3 can be trained end-to-end with machine-learned sequence processing model(s) 4.

[0140] Figure 7 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model development platform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0141] Model development platform 12 can provide one or more model libraries 13 containing building blocks for new models. Model libraries 13 can include one or more pretrained foundational models 13-1, which can provide a backbone of processing power acrossvarious tasks. Model libraries 13 can include one or more pre-trained expert models 13-2, which can be focused on performance in particular domains of expertise. Model libraries 13 can include various model primitives 13-3, which can provide low-level architectures or components (optionally pre-trained), which can be assembled in various arrangements as desired. Model primitives 13-3 can include a library of pre-trained adapters or LoRA modules that can adapt a baseline foundational model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like.

[0142] Model development platform 12 can receive selections of various model components 14. Model development platform 12 can pass selected model components 14 to a workbench 15 that combines selected model components 14 into a development model 16.

[0143] Workbench 15 can facilitate further refinement and adaptation of development model 16 by leveraging a number of different toolkits integrated with model development platform 12. For example, workbench 15 can facilitate alignment of the development model 16 with a desired performance profile on various tasks using a model alignment toolkit 17.

[0144] Model alignment toolkit 17 can provide a number of tools for causing development model 16 to generate outputs aligned with desired behavioral characteristics. Alignment can include increasing the accuracy, precision, recall, etc. of model outputs.Alignment can include enforcing output styles, schema, or other preferential characteristics of model outputs. Alignment can be general or domain-specific. For instance, a pre- trained foundational model 13-1 can begin with an initial level of performance across multiple domains. Alignment of the pre-trained foundational model 13-1 can include improving a performance in a particular domain of information or tasks (e g., even at the expense of performance in another domain of information or tasks).

[0145] Model alignment toolkit 17 can integrate one or more dataset(s) 17-1 for aligning development model 16. Curated dataset(s) 17-1 can include labeled or unlabeled training data. Dataset(s) 17-1 can be obtained from public domain datasets. Dataset(s) 17-1 can be obtained from private datasets associated with one or more developer system(s) for the alignment of bespoke machine-learned model(s) customized for private use-cases.

[0146] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e g., denoising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16.

[0147] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher- quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to finetune development model 16.

[0148] Prompt libraries 17-4 can include sets of inputs configured to induce behavior aligned with desired performance criteria. Prompt libraries 17-4 can include few-shot prompts (e.g., inputs providing examples of desired model outputs for prepending to a desired runtime query), chain-of-thought prompts (e.g., inputs providing step-by-step reasoning within the exemplars to facilitate thorough reasoning by the model), and the like.

[0149] Example prompts can be retrieved from an available repository of prompt libraries 17-4. Example prompts can be contributed by one or more developer systems using workbench 15.

[0150] In some implementations, pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts can include inputs that lack exemplars. Zero-shot prompts can be within a domain within a training dataset or outside of the training domain(s).

[0151] Prompt libraries 17-4 can include one or more prompt engineering tools. Prompt engineering tools can provide workflows for retrieving or learning optimized prompt values. Prompt engineering tools can facilitate directly learning prompt values (e.g., input element values) based on one or more training iterations. Workbench 15 can implement prompt engineering tools in development model 16.

[0152] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine- learned models. In this manner, for instance, a first model can process information about a task and output an input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0153] Prompt libraries 17-4 can include pipelines for context injection. For instance, a performance of development model 16 on a particular task can improve if provided withadditional context for performing the task. Prompt libraries 17-4 can include software components configured to identify desired context, retrieve the context from an external source (e.g., a database, a sensor, etc.), and add the context to the input prompt. Workbench 15 can implement context injection pipelines in development model 16.

[0154] Although various training examples described herein with respect to model development platform 12 refer to “‘pre-training” and “fine-tuning.” it is to be understood that model alignment toolkit 17 can generally support a wide variety of training techniques adapted for training a wide variety of machine-learned models. Example training techniques can correspond to the example training method 300 described above.

[0155] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance quality where appropriate. For instance, deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error. For instance, instead of autoregressively predicting the solution to a system of equations, a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool. The tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations. The output of the tool can be returned in response to the original query. In this manner, tool use can allow some example models to focus on the strengths of machine-learned models — e.g., understanding an intent in an unstructured request for a task — while augmenting the performance of the model by offloading certain tasks to a more focused tool for rote application of deterministic algorithms to a well-defined problem.

[0156] Model plugin toolkit 18 can include validation tools 18-1. Validation tools 18- 1 can include tools that can parse and confirm output(s) of a machine-learned model.Validation tools 18-1 can include engineered heuristics that establish certain thresholds applied to model outputs. For example, validation tools 18-1 can ground the outputs of machine-learned models to structured data sources (e.g., to mitigate “hallucinations”).

[0157] Model plugin toolkit 18 can include tooling packages 18-2 for implementing one or more tools that can include scripts or other executable code that can be executed alongside development model 16. Tooling packages 18-2 can include one or more inputs configured to cause machine-learned model(s) to implement the tools (e.g.. few-shot promptsthat induce a model to output tool calls in the proper syntax, etc.). Tooling packages 18-2 can include, for instance, fine-tuning training data for training a model to use a tool.

[0158] Model plugin toolkit 18 can include interfaces for calling external application programming interfaces (APIs) 18-3. For instance, in addition to or in lieu of implementing tool calls or tool code directly with development model 16, development model 16 can be aligned to output instructions that initiate API calls to send or obtain data via external systems.

[0159] Model plugin toolkit 18 can integrate with prompt libraries 17-4 to build a catalog of available tools for use with development model 16. For instance, a model can receive, in an input, a catalog of available tools, and the model can generate an output that selects a tool from the available tools and initiates a tool call for using the tool.

[0160] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardware acceleration 19-2 can facilitate the configuration of the model storage and execution formats to operate optimally on different hardware resources. For instance, hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memory requirements, etc. Tools for distillation 19-3 can provide for the training of lighter-weight models based on the knowledge encoded in development model 16. For instance, development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12. To obtain a lightweight model for running in resource-constrained environments, a smaller model can be a '‘student model’’ that learns to imitate development model 16 as a “teacher model.” In this manner, for instance, the investment in learning the parameters and configurations of development model 16 can be efficiently transferred to a smaller model for more efficient inference.

[0161] Workbench 15 can implement one, multiple, or none of the toolkits implemented in model development platform 12. Workbench 15 can output an output model 20 based on development model 16. Output model 20 can be a deployment version of development model 16. Output model 20 can be a development or training checkpoint of development model 16. Output model 20 can be a distilled, compressed, or otherwise optimized version of development model 16.

[0162] Figure 8 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example training flow can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 8 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 8 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0163] Initially, development model 16 can persist in an initial state as an initialized model 21. Development model 16 can be initialized with weight values. Initial weight values can be random or based on an initialization schema. Initial weight values can be based on prior pre-training for the same or for a different model.

[0164] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pretraining stage 22 can be implemented using one or more pre-training pipelines 17-2 over data from dataset(s) 17-1. Pre-training can be omitted, for example, if initialized model 21 is already pre-trained (e.g.. development model 16 contains, is, or is based on a pre-trained foundational model or an expert model).

[0165] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Pre-trained model 23 can be the initial state if development model 16 was already pre-trained. Pre-trained model 23 can undergo fine-tuning in a fine-tuning stage 24. Fine-tuning stage 24 can be implemented using one or more fine-tuning pipelines 17-3 over data from dataset(s) 17-1. Fine-tuning can be omitted, for example, if a pre-trained model has satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0166] Fine-tuned model 29 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Fine-tunedmodel 29 can be the initial state if development model 16 was already fine-tuned. Fine-tuned model 29 can undergo refinement with user feedback 26. For instance, refinement with user feedback 26 can include reinforcement learning, optionally based on human feedback from human users of fine-tuned model 25. As reinforcement learning can be a form of fine-tuning, it is to be understood that fine-tuning stage 24 can subsume the stage for refining with user feedback 26. Refinement with user feedback 26 can produce a refined model 27. Refined model 27 can be output to downstream system(s) 28 for deployment or further development.

[0167] In some implementations, computational optimization operations can be applied before, during, or after each stage. For instance, initialized model 21 can undergo computational optimization 29-1 (e.g., using computational optimization toolkit 19) before pre-training stage 22. Pre-trained model 23 can undergo computational optimization 29-2 (e.g., using computational optimization toolkit 19) before fine-tuning stage 24. Fine-tuned model 25 can undergo computational optimization 29-3 (e.g., using computational optimization toolkit 19) before refinement with user feedback 26. Refined model 27 can undergo computational optimization 29-4 (e.g., using computational optimization toolkit 19) before output to downstream system(s) 28. Computational optimization(s) 29-1, . . . . 29-4 can all be the same, all be different, or include at least some different optimization techniques.

[0168] Figure 9 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.). A model host 31 can receive machine-learned model(s) 1 . Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31.

[0169] Model host 31 can perform inference on behalf of one or more client(s) 32. Client(s) 32 can transmit an input request 33 to model host 31. Using input request 33, model host 31 can obtain input(s) 2 for input to machine-learned model(s) 1. Machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3. Using output(s) 3, model host 31 can return an output payload 34 for responding to input request 33 from client(s) 32. Output payload 34 can include or be based on output(s) 3.

[0170] Model host 31 can leverage various other resources and tools to augment the inference task. For instance, model host 31 can communicate with tool interfaces 35 to facilitate tool use by model instance(s) 31-1. Tool interfaces 35 can include local or remote APIs. Tool interfaces 35 can include integrated scripts or other software functionality. Modelhost 31 can engage online learning interface(s) 36 to facilitate ongoing improvements to machine-learned model(s) 1. For instance, online learning interface(s) 36 can be used within reinforcement learning loops to retrieve user feedback on inferences served by model host 31. Model host 31 can access runtime data source(s) 37 for augmenting input(s) 2 with additional contextual information. For instance, runtime data source(s) 37 can include a knowledge graph 37-1 that facilitates structured information retrieval for information associated with input request(s) 33 (e.g., a search engine service). Runtime data source(s) 37 can include public or private, external or local database(s) 37-2 that can store information associated with input request(s) 33 for augmenting input(s) 2. Runtime data source(s) 37 can include account data 37-3 which can be retrieved in association with a user account corresponding to a client 32 for customizing the behavior of model host 31 accordingly.

[0171] Model host 31 can be implemented by one or multiple computing devices or systems. Client(s) 2 can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31.

[0172] For example, model host 31 can operate on a server system that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network). Client device(s) can be end-user devices used by individuals. Client device(s) can be server systems that operate client(s) 32 to provide various functionality as a service to downstream end-user devices.

[0173] In some implementations, model host 31 can operate on the same device or system as client(s) 32. Model host 31 can be a machine-learning service that runs on-device to provide machine-learning functionality to one or multiple applications operating on a client device, which can include an application implementing client(s) 32. Model host 31 can be a part of the same application as client(s) 32. For instance, model host 31 can be a subroutine or method implemented by one part of an application, and client(s) 32 can be another subroutine or method that engages model host 31 to perform inference functions within the application. It is to be understood that model host 31 and client(s) 32 can have various different configurations.

[0174] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include weights or other model components that are stored in persistent storage, temporarily cached, or loaded into high-speed memory'. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include cachedintermediate states of active or inactive model(s) used to accelerate inference of those models. For instance, an inference session with a particular model may generate significant amounts of computational results that can be re-used for future inference runs (e.g., using a KV cache for transformer-based models). These computational results can be saved in association with that inference session so that session can be executed more efficiently when resumed.

[0175] Compute resource(s) 31-2 can include one or more processors (central processing units, graphical processing units, tensor processing units, machine-learning accelerators, etc.) connected to one or more memory' devices. Compute resource(s) 31-2 can include a dynamic pool of available resources shared with other processes. Compute resource(s) 31-2 can include memory devices large enough to fit an entire model instance in a single memory instance. Compute resource(s) 31-2 can also shard model instance(s) across multiple memory devices (e.g., using data parallelization or tensor parallelization, etc.). This can be done to increase parallelization or to execute a large model using multiple memory devices which individually might not be able to fit the entire model into memory.

[0176] Input request 33 can include data for input(s) 2. Model host 31 can process input request 33 to obtain input(s) 2. Input(s) 2 can be obtained directly from input request 33 or can be retrieved using input request 33. Input request 33 can be submitted to model host 31 via an API.

[0177] Model host 31 can perform inference over batches of input requests 33 in parallel. For instance, a model instance 31 -1 can be configured with an input structure that has a batch dimension. Separate input(s) 2 can be distributed across the batch dimension (e.g., rows of an array). The separate input(s) 2 can include completely different contexts. The separate input(s) 2 can be multiple inference steps of the same task. The separate input(s) 2 can be staggered in an input structure, such that any given inference cycle can be operating on different portions of the respective input(s) 2. In this manner, for instance, model host 31 can perform inference on the batch in parallel, such that output(s) 3 can also contain the batch dimension and return the inference results for the batched input(s) 2 in parallel. In this manner, for instance, batches of input request(s) 33 can be processed in parallel for higher throughput of output payload(s) 34.

[0178] Output payload 34 can include or be based on output(s) 3 from machine- learned model(s) 1. Model host 31 can process output(s) 3 to obtain output payload 34. This can include chaining multiple rounds of inference (e.g., iteratively, recursively, across thesame model(s) or different model(s)) to arrive at a final output for a task to be returned in output payload 34. Output payload 34 can be transmitted to client(s) 32 via an API.

[0179] Online learning interface(s) 36 can facilitate reinforcement learning of machine-learned model(s) 1. Online learning interface(s) 36 can facilitate reinforcement learning with human feedback (RLHF). Online learning interface(s) 36 can facilitate federated learning of machine-learned model(s) 1.

[0180] Model host 31 can access a library of pre-trained adapters or LoRA modules that can adapt a baseline model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like. For instance, model host 31 can receive an input request to load a customized model, and model host 31 can retrieve one or more components to adapt a baseline model to the custom profile. Model host 31 can determine that a particular functionality is needed for a particular task (e.g., based on an output of a model that preprocesses an input) and retrieve a pre-trained component accordingly.

[0181] Model host 31 can execute machine-learned model(s) 1 to perform inference for various tasks using various types of data. For example, various different input(s) 2 and output(s) 3 can be used for various different tasks. In some implementations, input(s) 2 can be or otherwise represent image data. Machine-learned model(s) 1 can process the image data to generate an output. As an example, machine-learned model(s) 1 can process the image data to generate an image recognition output (e.g., a recognition of the image data, a latent embedding of the image data, an encoded representation of the image data, a hash of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an image segmentation output. As another example, machine-learned model(s) 1 can process the image data to generate an image classification output. As another example, machine-learned model(s) 1 can process the image data to generate an image data modification output (e.g., an alteration of the image data, etc.). As another example, machine- learned model(s) 1 can process the image data to generate an encoded image data output (e.g., an encoded and / or compressed representation of the image data, etc.). As another example, machine-learned model(s) I can process the image data to generate an upscaled image data output. As another example, machine-learned model(s) 1 can process the image data to generate a prediction output.

[0182] In some implementations, the task is a computer vision task. In some cases, input(s) 2 includes pixel data for one or more images and the task is an image processing task. For example, the image processing task can be image classification, where the output isa set of scores, each score corresponding to a different object class and representing the likelihood that the one or more images depict an object belonging to the object class. The image processing task may be object detection, where the image processing output identifies one or more regions in the one or more images and, for each region, a likelihood that region depicts an object of interest. As another example, the image processing task can be image segmentation, where the image processing output defines, for each pixel in the one or more images, a respective likelihood for each category in a predetermined set of categories. For example, the set of categories can be foreground and background. As another example, the set of categories can be object classes. As another example, the image processing task can be depth estimation, where the image processing output defines, for each pixel in the one or more images, a respective depth value. As another example, the image processing task can be motion estimation, where the network input includes multiple images, and the image processing output defines, for each pixel of one of the input images, a motion of the scene depicted at the pixel between the images in the network input.

[0183] In some implementations, input(s) 2 can be or otherwise represent natural language data. Machine-learned model(s) 1 can process the natural language data to generate an output. As an example, machine-learned model(s) 1 can process the natural language data to generate a language encoding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a latent text embedding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a translation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a classification output. As another example, machine-learned model(s) 1 can process the natural language data to generate a textual segmentation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a semantic intent output. As another example, machine-learned model(s) 1 can process the natural language data to generate an upscaled text or natural language output (e.g., text or natural language data that is higher quality than the input text or natural language, etc.). As another example, machine-learned model(s) 1 can process the natural language data to generate a prediction output (e.g., one or more predicted next portions of natural language content).

[0184] In some implementations, input(s) 2 can be or otherwise represent speech data (e.g., data describing spoken natural language, such as audio data, textual data, etc.). Machine-learned model(s) 1 can process the speech data to generate an output. As an example, machine-learned model(s) 1 can process the speech data to generate a speechrecognition output. As another example, machine-learned model(s) 1 can process the speech data to generate a speech translation output. As another example, machine-learned model(s) 1 can process the speech data to generate a latent embedding output. As another example, machine-learned model(s) 1 can process the speech data to generate an encoded speech output (e.g., an encoded and / or compressed representation of the speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate an upscaled speech output (e.g., speech data that is higher quality than the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a textual representation output (e.g., a textual representation of the input speech data, etc.). As another example, machine-learned model(s) 1 can process the speech data to generate a prediction output.

[0185] In some implementations, input(s) 2 can be or otherwise represent latent encoding data (e.g., a latent space representation of an input, etc.). Machine-learned model(s) 1 can process the latent encoding data to generate an output. As an example, machine- learned model(s) 1 can process the latent encoding data to generate a recognition output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reconstruction output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a search output. As another example, machine- learned model(s) 1 can process the latent encoding data to generate a reclustering output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.

[0186] In some implementations, input(s) 2 can be or otherwise represent statistical data. Statistical data can be, represent, or otherwise include data computed and / or calculated from some other data source. Machine-learned model(s) 1 can process the statistical data to generate an output. As an example, machine-learned model(s) 1 can process the statistical data to generate a recognition output. As another example, machine-learned model(s) 1 can process the statistical data to generate a prediction output. As another example, machine- learned model(s) 1 can process the statistical data to generate a classification output. As another example, machine-learned model(s) 1 can process the statistical data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the statistical data to generate a visualization output. As another example, machine-learned model(s) 1 can process the statistical data to generate a diagnostic output.

[0187] In some implementations, input(s) 2 can be or otherwise represent sensor data. Machine-learned model(s) 1 can process the sensor data to generate an output. As anexample, machine-learned model(s) 1 can process the sensor data to generate a recognition output. As another example, machine-learned model(s) 1 can process the sensor data to generate a prediction output. As another example, machine-learned model(s) 1 can process the sensor data to generate a classification output. As another example, machine-learned model(s) 1 can process the sensor data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the sensor data to generate a visualization output. As another example, machine-learned model(s) 1 can process the sensor data to generate a diagnostic output. As another example, machine-learned model(s) 1 can process the sensor data to generate a detection output.

[0188] In some implementations, machine-learned model (s) 1 can be configured to perform a task that includes encoding input data for reliable and / or efficient transmission or storage (and / or corresponding decoding). For example, the task may be an audio compression task. The input may include audio data and the output may comprise compressed audio data. In another example, the input includes visual data (e.g. one or more images or videos), the output comprises compressed visual data, and the task is a visual data compression task. In another example, the task may comprise generating an embedding for input data (e.g. input audio or visual data). In some cases, the input includes audio data representing a spoken utterance and the task is a speech recognition task. The output may comprise a text output which is mapped to the spoken utterance. In some cases, the task comprises encrypting or decrypting input data. In some cases, the task comprises a microprocessor performance task, such as branch prediction or memory address translation.

[0189] In some implementations, the task is a generative task, and machine-learned model(s) 1 can be configured to output content generated in view of input(s) 2. For instance, input(s) 2 can be or otherwise represent data of one or more modalities that encodes context for generating additional content.

[0190] In some implementations, the task can be a text completion task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent textual data and to generate output(s) 3 that represent additional textual data that completes a textual sequence that includes input(s) 2. For instance, machine-learned model(s) 1 can be configured to generate output(s) 3 to complete a sentence, paragraph, or portion of text that follows from a portion of text represented by input(s) 2.

[0191] In some implementations, the task can be an instruction-following task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent instructions to perform a function and to generate output(s) 3 that advance a goal of satisfyingthe instruction function (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the instructions (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward accomplishing the requested functionality. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of performing a function. Multiple steps can be performed, with a final output being obtained that is responsive to the initial instructions.

[0192] In some implementations, the task can be a question answering task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent a question to answer and to generate output(s) 3 that advance a goal of returning an answer to the question (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine- learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward answering the question. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of obtaining an answer to the question (e g., querying a database, performing a computation, executing a script, etc.). Multiple steps can be performed, with a final output being obtained that is responsive to the question.

[0193] In some implementations, the task can be an image generation task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of image content. The context can include text data, image data, audio data, etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent image data that depicts imagery7related to the context. For instance, machine-learned model(s) 1 can be configured to generate pixel data of an image. Values for channel (s) associated with the pixels in the pixel data can be selected based on the context (e.g., based on a probability determined based on the context).

[0194] In some implementations, the task can be an audio generation task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of audio content. The context can include text data, image data, audio data, etc. Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent audio data related to the context. For instance, machine-learned model (s) 1 can be configured to generate waveform data in the form of an image (e.g., a spectrogram). Values for channel(s) associated with pixels of the image can be selected based on the context. Machine- learned model(s) 1 can be configured to generate waveform data in the form of a sequence of discrete samples of a continuous waveform. Values of the sequence can be selected based on the context (e.g., based on a probability7determined based on the context).

[0195] In some implementations, the task can be a data generation task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of data (e.g., data from various data domains, such as sensor data, image data, multimodal data, statistical data, etc.). The desired data can be, for instance, synthetic data for training other machine-learned models. The context can include arbitrary data type(s). Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent data that aligns with the desired data. For instance, machine-learned model (s) 1 can be configured to generate data values for populating a dataset. Values for the data object(s) can be selected based on the context (e.g., based on a probability7determined based on the context).

[0196] Figure 10 is a block diagram of an example networked computing system that can perform aspects of example implementations of the present disclosure. The system can include a number of computing devices and systems that are communicatively coupled over a network 49. An example computing device 50 is described to provide an example of a computing device that can perfomi any aspect of the present disclosure (e.g., implementing model host 31. client(s) 32, or both). An example server computing system 60 is described as an example of a server computing system that can perform any aspect of the presentdisclosure (e.g., implementing model host 31, client(s) 32, or both). Computing device 50 and server computing system(s) 60 can cooperatively interact (e.g., over network 49) to perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Model development platform system 70 is an example system that can host or serve model development platform(s) 12 for development of machined earned models. Third-party' system(s) 80 are example system(s) with which any of computing device 50, server computing system(s) 60, or model development platform system(s) 70 can interact in the performance of various aspects of the present disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.).

[0197] Network 49 can be any ty pe of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof and can include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of yvired or yvireless connection, using a yvide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), or protection schemes (e.g., VPN. secure HTTP, SSL). Network 49 can also be implemented via a system bus. For instance, one or more devices or systems of Figure 10 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0198] Computing device 50 can be any type of computing device, such as, for example, a personal computing device (e.g.. laptop or desktop), a mobile computing device (e.g., smartphone or tablet), a gaming console or controller, a wearable computing device, an embedded computing device, a server computing device, a virtual machine operating on a host device, or any other type of computing device. Computing device 50 can be a client computing device. Computing device 50 can be an end-user computing device. Computing device 50 can be a computing device of a service provided that provides a service to an end user (who may use another computing device to interact yvith computing device 50).

[0199] Computing device 50 can include one or more processors 51 and a memory 52. Processor(s) 51 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 52 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory' devices, magnetic disks, etc., and combinations thereof. Memory 52 can store data 53 and instructions 54 which can be executed by processor(s) 51 to cause computing device 50 to perform operations. The operations can implement any one ormultiple features described herein. The operations can implement example methods and techniques described herein.

[0200] Computing device 50 can also include one or more input components that receive user input. For example, a user input component can be a touch-sensitive component (e.g., a touch-sensitive display screen or a touch pad) that is sensitive to the touch of a user input object (e.g., a finger or a stylus). The touch-sensitive component can serve to implement a virtual keyboard. Other example user input components include a microphone, camera, LIDAR, a physical keyboard or other buttons, or other means by which a user can provide user input.

[0201] Computing device 50 can store or include one or more machine-learned models 55. Machine-learned models 55 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 55 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 55 can be received from server computing system(s) 60, model development platform system 70, third party system(s) 80 (e.g., an application distribution platform), or developed locally on computing device 50. Machine-learned model(s) 55 can be loaded into memory 52 and used or otherwise implemented by processor(s) 51. Computing device 50 can implement multiple parallel instances of machine-learned model (s) 55.

[0202] Server computing system(s) 60 can include one or more processors 61 and a memory 62. Processor(s) 61 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 62 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 62 can store data 63 and instructions 64 which can be executed by processor(s) 61 to cause server computing system(s) 60 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein.

[0203] In some implementations, server computing system 60 includes or is otherwise implemented by one or multiple server computing devices. In instances in which server computing system 60 includes multiple server computing devices, such server computing devices can operate according to sequential computing architectures, parallel computing architectures, or some combination thereof.

[0204] Server computing system 60 can store or otherwise include one or more machine-learned models 65. Machine-learned model(s) 65 can be the same as or different from machine-learned model(s) 55. Machine-learned models 65 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 65 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 65 can be received from computing device 50, model development platform system 70, third party system(s) 80. or developed locally on server computing system(s) 60. Machine-learned model(s) 65 can be loaded into memory 62 and used or otherwise implemented by processor(s) 61. Server computing system(s) 60 can implement multiple parallel instances of machine-learned model(s) 65.

[0205] In an example configuration, machine-learned models 65 can be included in or otherwise stored and implemented by server computing system 60 to establish a client-server relationship with computing device 50 for serving model inferences. For instance, server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50. For instance, machine-learned models 65 can be implemented by server computing system 60 as a portion of a web service (e.g., remote machine-learned model hosting service, such as an online interface for performing machine-learned model operations over a network on server computing system(s) 60). For instance, server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection. For instance, computing device 50 can be a workstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e.g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50. Machine-learned models 65 can work cooperatively or interoperatively with machine- learned models 55 on computing device 50 to perform various tasks.

[0206] Model development platform system(s) 70 can include one or more processors 71 and a memory 72. Processor(s) 71 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 72 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory7devices, magnetic disks, etc., and combinations thereof. Memory 72 can store data 73 and instructions 74 which can be executed by processor(s) 71 to cause model development platform system(s) 70 to perform operations. The operations can implement any one or multiple features described herein. The operationscan implement example methods and techniques described herein. Example operations include the functionality described herein with respect to model development platform 12. This and other functionality can be implemented by developer tool(s) 75.

[0207] Third-party system(s) 80 can include one or more processors 81 and a memory 82. Processor(s) 81 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 82 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory' devices, magnetic disks, etc., and combinations thereof. Memory 82 can store data 83 and instructions 84 which can be executed by processor(s) 81 to cause third-party system(s) 80 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein with respect to tools and other external resources called when training or performing inference with machine-learned model(s) 1, 4. 16. 20, 55, 65, etc. (e.g., third-party resource(s) 85).

[0208] Figure 10 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update / train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or more techniques described herein with respect to model alignment toolkit 17. In this manner, for instance, computing system 50 or server computing system(s) 60 can develop, update / train, or refine machine-learned models based on local datasets (e.g., for model personalization / customization, as permitted by user data preference selections).

[0209] Figure 11 is a block diagram of an example computing device 98 that performs according to example embodiments of the present disclosure. Computing device 98 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 98 can include a number of applications (e.g., applications 1 through N). Each application can contain its own machine learning library and machine- learned model(s). For example, each application can include a machine-learned model.Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. As illustrated in Figure 11, each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application can communicate with each device component using an API (e.g., a public API). In some implementations, the API used by each application is specific to that application.

[0210] Figure 12 is a block diagram of an example computing device 99 that performs according to example embodiments of the present disclosure. Computing device 99 can be the same as or different from computing device 98. Computing device 99 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 99 can include a number of applications (e.g., applications 1 through N). Each application can be in communication with a central intelligence layer. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. In some implementations, each application can communicate with the central intelligence layer (and model(s) stored therein) using an API (e.g., a common API across all applications).

[0211] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in Figure 12, a respective machine-learned model can be provided for each application and managed by the central intelligence layer. In other implementations, two or more applications can share a single machine-learned model. For example, in some implementations, the central intelligence layer can provide a single model for all of the applications. In some implementations, the central intelligence layer is included within or otherwise implemented by an operating system of computing device 99.

[0212] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in Figure 12, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0213] The technology discussed herein makes reference to servers, databases, software applications, and other computer-based systems, as well as actions taken andinformation sent to and from such systems. The inherent flexibility of computer-based systems allows for a great variety of possible configurations, combinations, and divisions of tasks and functionality between and among components. For instance, processes discussed herein can be implemented using a single device or component or multiple devices or components working in combination. Databases and applications can be implemented on a single system or distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0214] While the present subject matter has been described in detail with respect to various specific example embodiments thereof, each example is provided by way of explanation, not limitation of the disclosure. Those skilled in the art, upon attaining an understanding of the foregoing, can readily produce alterations to, variations of, and equivalents to such embodiments. Accordingly, the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that the present disclosure cover such alterations, variations, and equivalents.

[0215] Aspects of the disclosure have been described in terms of illustrative embodiments thereof. Any and all features in the following claims can be combined or rearranged in any way possible, including combinations of claims not explicitly enumerated in combination together, as the example claim dependencies listed herein should not be read as limiting the scope of possible combinations of features disclosed herein. Accordingly, the scope of the present disclosure is by way of example rather than by w ay of limitation, and the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. Moreover, terms are described herein using lists of example elements joined by conjunctions such as “and,” “or,” “but,” etc. It should be understood that such conjunctions are provided for explanatory purposes only. Clauses and other sequences of items joined by a particular conjunction such as “or,” for example, can refer to “and / or,” “at least one of’, “any combination of’ example elements listed therein, etc. Terms such as “based on” should be understood as “based at least in part on.”

[0216] The term “can” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X can perform Y” should be understood asindicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every’ instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

[0217] The term “may'’ should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X may perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

Claims

WHAT IS CLAIMED IS:

1. A computer-implemented method, the method comprising: obtaining, by a computing system comprising one or more computing devices, observation data associated with a web browser application, wherein the observation data was captured by a browser agent extension associated with the web browser application; generating, by the computing system, a model input based on the observation data; generating, by the computing system and based on processing the model input using a machine-learned sequence processing model, a model output; outputting, by the computing system, and based on the model output, a browser action that specifies an action to be performed relative to the web browser application; and causing, by the computing system, the browser action to be performed in the web browser application.

2. The computer-implemented method of claim 1, further comprising: automatically iteratively performing the method of claim 1 until a stop condition is met, wherein the stop condition comprises a user intervention; receiving a clarifying user input; and after receiving the clarifying user input, resuming iterative performance of the method of claim 1, wherein at least one model input in the iterative performance is based on the clarifying user input.

3. The computer-implemented method of any preceding claim, wherein the observation data comprises one or more images that depict a user interface of the web browser application.

4. The computer-implemented method of claim 3, wherein the observation data further comprises textual summaries of one or more older images that depict older states of the user interface of the web browser application.

5. The computer-implemented method of any preceding claim, wherein obtaining the observation data further comprises annotating the observation data with one or more identifiers that indicate interactive elements included in a web document being browsed by the web browser application.

6. The computer-implemented method of any preceding claim, wherein the method further comprises receiving input data associated with a user, and wherein generating the model input comprises generating the model input based on both the observation data and the input data.

7. The computer-implemented method of claim 6, wherein the input data comprises speech data or textual data input by the user, wherein the input data specifies a requested task, and wherein the browser action advances the web browser application toward completion of the requested task.

8. The computer-implemented method of any preceding claim, wherein the browser action is contained within a pre-defined action space comprising a plurality of available actions that are able to be performed in the web browser application.

9. The computer-implemented method of any preceding claim, wherein the browser action comprises a clarify action that requests a user to provide clarifying user input.

10. The computer-implemented method of any preceding claim, wherein the sequence processing model is configured to perform tool use in which the sequence processing model makes calls to one or more external tools to obtain additional contextual information from the one or more external tools.

11. The computer-implemented method of any preceding claim, wherein: outputting, by the computing system, and based on the model output, the browser action comprises: parsing the model output into a thought output and an action output; and generating the browser action based on the action output; and the method further comprises causing, by the computing system, a thought output portion of the model output to be provided as an output to a user.

12. The computer-implemented method of any preceding claim, wherein the browser agent extension comprises actuation logic that performs the browser action in the web browser application.

13. The computer-implemented method of any preceding claim, wherein: the method further comprises maintaining, by the computing system, current session data associated with a current browsing session, andgenerating, by the computing system, the model input comprises generating, by the game companion agent, the model input based at least in part on the current session data.

14. The computer-implemented method of any preceding claim, wherein: the method further comprises maintaining, by the computing system, prior session data associated with one or more prior browsing sessions, and generating, by the computing system, the model input comprises generating, by the computing system, the model input based at least in part on the prior session data.

15. The computer-implemented method of any preceding claim, wherein the sequence processing model was conditioned on one or more workflow demonstrations that demonstrate sequences of interactions with the web browser application.

16. The computer-implemented method of any preceding claim, wherein the sequence processing model was trained on one or more workflow demonstrations that demonstrate sequences of interactions with the web browser application.

17. The computer-implemented method of any preceding claim, wherein the method further comprises: causing, by the computing system, a dedicated user interface window, panel, or element to be rendered within a user interface of the web browser application, wherein the dedicated user interface window, panel, or element depicts interactions, activities, or other aspects of the browser agent extension.

18. The computer-implemented method of any preceding claim, further comprising: injecting HTML code into the web browser application to cause the web browser application to visualize one or more visual elements within a user interface of the web browser application, wherein the visual elements depict performance of the browser action.

19. The computer-implemented method of any preceding claim, further comprising: visualizing an agent cursor associated with the browser agent extension.

20. The computer-implemented method of any preceding claim, further comprising: visualizing a heads up display of keystrokes associated with the browser agent extension.

21. The computer-implemented method of any preceding claim, wherein the observation data comprises a sequence of screenshots or pixel data capturing a visual state of the web browser application, and wherein generating the model input comprises processing the screenshots or pixel data using a multimodal vision model.

22. The computer-implemented method of any preceding claim, wherein the browser agent extension interacts with the web browser application via an application programming interface to capture the observation data and inject the browser action.

23. The computer-implemented method of any preceding claim, wherein the web browser application and the browser agent extension are executed on a remote server computing system distinct from a client device associated with a user, and wherein causing the browser action to be performed comprises executing the browser action within the web browser application on the remote server computing system.

24. The computer implemented method of claim 23, further comprising: generating, by the remote server computing system, a visual stream of the user interface of the web browser application; and transmitting the visual stream to the client device for display to the user.

25. The computer-implemented method of any of claims 1-22, wherein the browser agent extension operates on a client device associated with a user to capture the observation data.

26. The computer-implemented method of any of claims 1-22, wherein the web browser application comprises a headless browser instance executing without rendering a graphical user interface on a display device, and wherein the browser agent extension monitors a programmatic state of the headless browser instance to generate the observation data.

27. A computing system, comprising: a browser agent extension configured to collect observations from a web browser application; and an action service; wherein the browser agent extension is configured to collect the observation data associated with the w eb brow ser application and to provide the observation data to the action service; wherein the action service is configured to obtain the observation data, generate a model input based on the observation data, process the model input with a machine-learned sequence processing model to generate a model output, generate a browser action based on the model output, and provide the browser action to the browser agent extension; andwherein the browser agent extension is configured to cause performance of the browser action within the web browser application.

28. The computing system of claim 27, wherein the browser agent extension is integrated into the web browser application.

29. The computing system of claim 27, wherein the browser agent extension is executed on a server system.

30. One or more non-transitory computer-readable media that collectively store the browser agent extension described in claim 27.

Images