Embedding system for computer-implemented agents

Abstract

Provided are systems and methods that generate and / or leverage a learned embedding space for computer-implemented agents. In particular, an embedding generation system can generate execution rollouts which can include or represent logs of computer-implemented agents performing various tasks, for example capturing detailed sequences of agent actions and outcomes. The embedding generation system can then use these execution rollouts to learn agent embeddings within an agent embedding space. These learned embeddings can then be leveraged to perform dynamic optimization of agent-task pairings, allowing for the selection of agents based on their demonstrated behaviors and capabilities, thus enhancing efficiency and effectiveness in task execution.

Description

EMBEDDING SYSTEM FOR COMPUTER-IMPLEMENTED AGENTSRELATED APPLICATIONS

[0001] This application claims priority to and the benefit of United States Provisional Patent Application Number 63 / 717,698, filed November 7, 2024, and titled Embedding System for Computer-Implemented Agents. United States Provisional Patent Application Number 63 / 717,698 is hereby incorporated by reference in its 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 systems and methods that generate and / or leverage a learned embedding space for computer- implemented agents.BACKGROUND

[0003] A computer can receive input(s). The computer can execute instructions to process the input(s) to generate output(s) using a parameterized model. The computer can obtain feedback on its performance in generating the outputs with the model. The computer can generate feedback by evaluating its performance. The computer can receive feedback from an external source. The computer can update parameters of the model based on the feedback to improve its performance. In this manner, the computer can iteratively ‘leam’’ to generate the desired outputs. The resulting model is often referred to as a machine-learned model.SUMMARY

[0004] 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.

[0005] One general aspect includes a method for generating a learned embedding space for computer-implemented agents. The method includes obtaining, by a computingsystem which may include one or more computing devices, a task dataset which may include data descriptive of a plurality’ of tasks. The method includes generating, by the computing system, a plurality of execution rollouts for a plurality7of computer-implemented agents based on the task dataset, where each of the plurality of execution rollouts may include a series of one or more outputs generated by one of the computer-implemented agents when performing one of the plurality of tasks. The method includes generating, by the computing system, a plurality of agent embeddings for the plurality7of computer-implemented agents, where the agent embedding for each computer-implemented agent is learned based on one or more of the execution rollouts that are associated with the computer-implemented agent. The method includes providing, by the computing system, the plurality of agent embeddings for the plurality of computer-implemented agents as an output. 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.

[0006] Example implementations may include any combination of one or more of the following features. The method of any preceding claim, where at least some of the plurality of execution rollouts may include a series of tuples which may include a state, an action, and a result. At least some of the plurality of execution rollouts may exhibit multi-step reasoning. At least some of the plurality7of execution rollouts may exhibit tool use. Generating at least some of the plurality of execution rollouts may include executing the computer-implemented agents in a sandbox computational environment. Generating at least some of the plurality of execution rollouts may include selecting one of the plurality7of computer-implemented agents and one of the plurality7of tasks at random. Generating at least some of the plurality7of execution rollouts may include selecting one of the plurality7of computer-implemented agents and one of the plurality7of tasks based on metadata that indicates that the selected agent is configured for performing the selected task. Generating, by the computing system, the plurality of agent embeddings for the plurality7of computer-implemented agents may7include training an autoencoder model on the plurality of execution rollouts, where the autoencoder model may include an encoder that leams to process an input execution rollout to generate an embedding and a decoder that leams to process the embedding to reconstruct the input execution rollout. Generating, by the computing system, the plurality of agent embeddings for the plurality7of computer-implemented agents may include performing a triplet training algorithm to generate the plurality of agent embeddings, where the triplet training algorithm trains an embedding generation model to minimize a distance between embeddings generated for input execution rollouts associated with the same computer-implemented agent and tomaximize a distance between embeddings generated for input execution rollouts associated with different computer-implemented agents. Generating, by the computing system, the plurality of agent embeddings for the plurality of computer-implemented agents may include generating a plurality of intermediate embeddings for the plurality of execution rollouts and, for each computer-implemented agent, averaging the intermediate embeddings respectively associated with the execution rollouts that are performed by such computer-implemented agent to generate the agent embedding for such computer-implemented agent. The method further may include generating a plurality of task embeddings respectively for the plurality’ of tasks. The method may include: receiving an input query: generating a query embedding for the input query; and identifying one or more of the computer-implemented agents as responsive to the input query based on a comparison of the query embedding to at least some of the plurality of agent embeddings. The input query may be descriptive of a new task. Identifying the one or more of the computer-implemented agents as responsive to the input query may include directly comparing the query embedding to the at least some of the plurality of agent embeddings. Providing the embeddings as an output can include providing the embeddings in a format that is searchable and / or queryable.

[0007] One general aspect is directed to one or more non-transitory computer- readable media that collectively store a plurality of agent embeddings that have been generated as described herein. The media may include computer-executable instructions for identifying the one or more of the computer-implemented agents as responsive to the input query’ which may include evaluating a mapping from the query embedding to the learned embedding space containing the plurality of agent embeddings for the plurality of computer- implemented agents. Implementations of the described techniques may include hardware, a method or process, or computer software on a computer-accessible medium.

[0008] One general aspect includes a computing system. The computing system includes one or more processors; and one or more non-transitory computer-readable media that collectively store: a plurality of agent embeddings respectively associated with a plurality of computer-implemented agents; and instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations may include: receiving an input query; generating a query’ embedding for the input query’; and identifying one or more of the computer-implemented agents as responsive to the input query based on a comparison of the query embedding to at least some of the plurality of agent embeddings. Other embodiments of this aspect include corresponding computer systems, apparatus, andcomputer programs recorded on one or more computer storage devices, each configured to perform the actions of the methods.

[0009] Example implementations can include any combination of one or more of the following features. The operations may further include providing the input query for processing by at least one of the one or more computer-implemented agents that were identified as responsive to the input query. The input query nay include one or more of textual content, audio content, or image content. The at least one of the one or more computer-implemented agents may generate a response to the input query. The response may include one or more of textual content, audio content, or image content. The operations may further include processing the input query with the at least one of the one or more computer- implemented agents that were identified as responsive to the input query. The input query may include a description of a first agent. Identifying the one or more computer-implemented agents as responsive to the input query may include identifying the one or more computer- implemented agents as being similar to the first agent.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 depicts a block diagram of an example technique for generating agent embeddings for computer-implemented agents according to example implementations of aspects of the present disclosure.

[0011] Figure 2 depicts a block diagram of an example auto-encoder-based embedding generation system according to example implementations of aspects of the present disclosure.

[0012] Figure 3 depicts a block diagram of an example triplet training approach according to example implementations of aspects of the present disclosure.

[0013] Figure 4 depicts block diagram of an example contrastive training approach according to example implementations of aspects of the present disclosure.

[0014] Figure 5 depicts a block diagram of an example approach for generating aggregated embeddings according to example implementations of aspects of the present disclosure.

[0015] Figure 6 depicts a block diagram of an example approach for generating agent and task embeddings in a shared embedding space according to example implementations of aspects of the present disclosure.

[0016] Figure 7 depicts a block diagram of an example agent search and / or routing system according to example implementations of aspects of the present disclosure.

[0017] Figure 8 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;

[0018] Figure 9 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;

[0019] Figure 10 is a block diagram of an example sequence processing model according to example implementations of aspects of the present disclosure;

[0020] Figure 11 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;

[0021] Figure 12 is a block diagram of an example model development platform according to example implementations of aspects of the present disclosure;

[0022] Figure 13 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;

[0023] Figure 14 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;

[0024] Figure 15 is a block diagram of an example networked computing system according to example implementations of aspects of the present disclosure;

[0025] Figure 16 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure; and

[0026] Figure 17 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure.DETAILED DESCRIPTION

[0027] In the field of artificial intelligence (Al), specifically in the domain of computer-implemented agents, a significant technical problem arises from the inefficiencies in task execution and resource allocation. Traditional systems typically employ static methods for task execution, which do not adapt to the varying demands of different tasks (e.g., different subtasks within a larger task). As such, traditional systems often struggle to achieve the optimal allocation of computational resources to computational tasks in adynamic environment, leading to suboptimal performance, increased execution times, and / or higher operational costs.

[0028] In particular, existing approaches often fail to effectively utilize the capabilities of different agents, resulting in a mismatch between an agent’s specific strengths and the tasks they are assigned. This inefficiency is compounded in environments where multiple agents with varying capabilities are available, as the static allocation systems do not account for the optimal pairing of agent capabilities with specific subtasks.

[0029] In view of these challenges, example aspects of the present disclosure are directed to systems and methods that generate and / or leverage a learned embedding space for computer-implemented agents. In particular, an embedding generation system can generate execution rollouts which can include or represent logs of computer-implemented agents performing various tasks, for example capturing detailed sequences of agent actions and outcomes. The embedding generation system can then use these execution rollouts to learn agent embeddings within an agent embedding space. These learned embeddings can then be leveraged to perform dynamic optimization of agent-task pairings, allowing for the selection of agents based on their demonstrated behaviors and capabilities, thus enhancing efficiency and effectiveness in task execution.

[0030] More particularly, an example embedding generation system can obtain a task dataset comprising data descriptive of a plurality of tasks. As one example, the task dataset can be assembled from a variety of sources. As one example, academic benchmarks such as BIG Bench, MMLU, or MMMU can provide standardized tasks which are often used to evaluate the capabilities of machine learning models across a range of cognitive tasks from problem-solving to understanding language. As another example, live benchmarks, such as LMSys, AgentBench, and APIBench, offer real-time data that reflect current challenges and scenarios that agents might encounter in practical applications.

[0031] Additionally or alternatively, the task dataset can also be sourced from live data and / or usage logs associated with a deployed production system. As one example, the task dataset can include data extracted from or otherwise associated with live or ongoing traffic handled by a deployed production system that contains one or multiple agents that are used to respond to queries, requests, or other inputs that contain various tasks. As another example, data logs which capture the most relevant task queries encountered in production systems can be obtained and added to the task dataset. These logs not only provide a direct insight into the types of tasks that are most commonly requested by users but may also include metadata about the success of the task completions, which can be useful forunderstanding the effectiveness of different agents. This metadata might include, for example, the time taken to complete the task, the accuracy of the task outcome, user satisfaction scores, and / or various other information such as whether certain tools were invoked or called (e.g., in the context of an agent-based tool use framework). This approach allows for a dynamically updating dataset that evolves with the changing conditions of the real-world environment in which the agents operate.

[0032] Following the collection of task data, the embedding generation system can generate execution rollouts for multiple computer-implemented agents based on this dataset. An execution rollout can include a series of outputs (e.g., intermediate outputs and / or a final output) generated by an agent as it attempts to perform a task. These outputs could be simple direct answers or can include complex multi-step reasoning, depending on the nature of the task and the capabilities of the agent.

[0033] In some implementations, to implement this concept, agents are indexed and then queried with arbitrary tasks from the collected task dataset, and the process of solving the task step-by-step is logged and stored in the form of an execution rollout. For example, agents might be randomly paired with tasks, prompting them to solve these tasks in various ways. In another example, agents can be paired with tasks using heuristics which maximize “coverage’' of all possible pairs of agents and tasks. Some agents might provide a direct answer immediately, which is then logged as part of the rollout. Alternatively, an agent may engage in a multi-step reasoning process, proposing solutions step-by-step, utilizing various tools during the process, and / or returning intermediate results and answers.

[0034] In some implementations, an execution rollout can be stored as a series of tuples like [state, action, (result), state, action, (result), ...], where both the state and action components may include complex structures such as tool identifiers, arguments, and / or answers, in addition to the usual text tokens that form the reasoning flow.

[0035] Thus, in some implementations, the execution rollouts can exhibit multi-step reasoning where an agent proposes solving a problem step-by-step, calling upon various tools in the process and returning intermediate results and answers. This is particularly useful in complex tasks that require nuanced decision-making and problem-solving capabilities.

[0036] Moreover, the execution rollouts can also showcase tool use, where agents utilize specific tools to aid in task resolution. For example, in a multi-step reasoning flow, the agent might use a language translation tool to understand user queries in different languages or employ a calculator for tasks requiring numerical computations. The use of tools can belogged as part of the execution rollout, capturing details such as tool IDs, arguments used, and the answers provided by these tools.

[0037] In some implementations, the agent might run through the task multiple times, especially if there is some stochastic behavior in the agent’s configuration, such as temperature sampling or tool failures. Multiple executions improves the robustness and reliability of the resulting execution rollout.

[0038] In some implementations, the execution of these rollouts can occur in a sandbox computational environment. This approach allows agents to perform tasks in a controlled setting where actions do not have real-world consequences. For instance, a restaurant booking agent might go through the motions of booking a table without actually making a real reservation. This sandboxing can be useful for testing and refining agent behaviors without the risks associated with real-world operations.

[0039] Furthermore, in some implementations, generating the execution rollouts can include randomly selecting one of the computer-implemented agents and one of the tasks from the dataset. This random sampling ensures a diverse range of interactions and behaviors are captured, contributing to a robust set of data for training and refining the agent embeddings. Alternatively, the selection of agents and tasks can be based on metadata that indicates a particular agent is well-suited for a specific task. This targeted approach can be beneficial for testing specific agent capabilities and ensuring that agents are operating within their areas of strength.

[0040] Having generated a plurality of execution rollouts, the embedding generation system can use these rollouts to generate a plurality of agent embeddings, where each agent’s embedding is learned based on the execution rollouts associated with that agent. These embeddings can effectively capture the unique behaviors, capabilities, and / or strategies of different agents when performing tasks. For example, two agents might achieve the same end result, but the paths they take could be vastly different, which would be reflected in their respective embeddings.

[0041] One possible example technique for generating these agent embeddings can include training an autoencoder model on the execution rollouts. In this approach, an encoder component of the autoencoder learns to process an input execution rollout to create an embedding. Subsequently, a decoder component attempts to reconstruct the input execution rollout from this embedding. This method ensures that the embeddings retain significant information about the rollout that can be used to differentiate between the behaviors of different agents effectively.

[0042] Another example technique for generating the agent embeddings can include using a triplet training algorithm. This algorithm is designed to minimize the distance between embeddings of execution rollouts associated with the same agent and maximize the distance between embeddings of rollouts associated with different agents. This approach is particularly effective in environments where agents perform similar tasks but in different ways, as it enhances the ability to distinguish between agents based on their task execution strategies.

[0043] Additionally, in some implementations, the embedding generation system can generate a plurality of intermediate embeddings for the execution rollouts (e.g., generate an intermediate embedding for each specific rollout). Then, the embedding generation system can average the intermediate embeddings that are respectively associated with execution rollouts performed by each agent to generate the agent’s final embedding. This averaging process helps in creating a more generalized representation of an agent’s behavior across multiple tasks.

[0044] Furthermore, in some implementations, each execution rollout might be split into multiple subsequences of [state, action, (result)], and embeddings could be learned for these subsequences. These intermediate embeddings are then aggregated to form the final agent embedding. This approach allows for a detailed analysis at each step of the task execution, providing a more granular view of the agent's operational methodology.

[0045] Thus, to determine general task-agnostic agent embeddings, the system might average the embeddings from multiple task executions by the same agent. This averaging process helps in assessing the overall capability and behavior pattern of an agent across various tasks, rather than in specific instances.

[0046] The system can also use metadata about agents to refine the embedding process. For instance, agents might be grouped and compared based on shared characteristics such as model type, version, or training data sources. This metadata can help in predicting the suitability of an agent for particular tasks, even before it is tested directly.

[0047] Having generated the agent embeddings, the system can provide these embeddings as a output. For example, the output can be a searchable and / or queryable output such as a database, for example facilitating an agent retrieval system to match agents with tasks they are suited for. This can be particularly useful in scenarios where a user or system needs to quickly determine which agent is best for a specific task.

[0048] In particular, this output (e.g., embedding database) is useful as it allows users or systems (e.g., other autonomous agents) to efficiently locate and utilize the optimal agentfor specific tasks based on their unique embeddings. For instance, if a human user or a computerized system (e.g.. computer-implemented agent) needs an agent specialized in data analysis, the agent retrieval system can quickly identify agents whose embeddings match this requirement.

[0049] In addition to agent embeddings, some example implementations can further include generating task embeddings for a variety’ of tasks. These task embeddings can be used to enhance the matching process between tasks and agents. By having a detailed representation of both agents and tasks in a shared or correlated embedding space, the system can more accurately pair them, ensuring that the selected agent is the best fit for the task at hand.

[0050] As one example, upon receiving an input query from a user, an agent retrieval system can generate a query embedding. This query embedding is then used to identify one or more computer-implemented agents that are best suited to respond to the query. This process can include comparing the query embedding directly to the agent embeddings stored in the system. Additionally or alternatively, the identification process can also include evaluating a mapping (e.g.. look-up table, hash function, etc.) from the query embedding to the learned embedding space containing the agent embeddings. For example, if a user queries about setting up an e-commerce website, the system will compare this query embedding to those of agents known for their proficiency in e-commerce platforms and suggest the most relevant agents.

[0051] In some implementations, at runtime, when a user provides a specific task, the agent retrieval system can access these precomputed task embeddings to retrieve the most similar tasks available in the vector database. The retrieved task may be linked with an agent that has performed best on such task. This retrieval process not only considers the similarity of tasks but also other criteria such as the likelihood of task success and cost-effectiveness of using one agent over another.

[0052] Additionally or alternatively, the system can build a task embedding space that maps directly to the agent embedding space. This mapping allows for a seamless transition from task identification to agent selection, streamlining the process of matching tasks with the best-suited agents. An embedding generation system might look at existing tasks and agents to build this function, or it might create a separate task embedding space and then generate a lookup table for mapping canonical tasks to specific agents.

[0053] Overall, the present disclosure provides a comprehensive system for embedding and retrieving agents based on their performance and capabilities, facilitating amore efficient and accurate selection process for various tasks. This system not only captures the unique behaviors and strategies of agents but may also adapt to new queries and tasks dynamically, ensuring relevance and precision in agent selection.

[0054] To provide an example of the proposed techniques, multiple computer- implemented agents may be capable of performing the same high-level task, but differ in their efficiency and method of task execution. For example, while one agent might use a textbased input to decompose a task, another might use multimodal inputs, leading to variations in performance. The disclosed technology can leverage an embedding mechanism to surface these differences. By comparing the embeddings of these agents, the system can recommend the most suitable agent for the task at hand. This capability not only potentially reduces the cost associated with performing the task but can also enhance the quality of the task execution, as the selected agent is optimally matched to the task requirements based on its demonstrated capabilities and methods.

[0055] The systems and methods of the present disclosure provide a number of technical effects and benefits. One example technical effect of the proposed technology is its ability to dynamically optimize resource allocation during task execution by computer- implemented agents. This technology specifically addresses inefficiencies in computational resource usage, which is a central aspect in the performance of Al systems. By intelligently allocating resources based on the demands of different subtasks, the technology- not only reduces the computational load but also decreases the execution time, directly contributing to the enhancement of system performance.

[0056] Another technical effect of the proposed technology is the capability to tailor agent-task pairings according to the specific strengths and capabilities of different agents. This adaptation ensures that each component of a task is handled by an agent most suited for that particular function, leading to more performant task execution. Such a strategic deployment of agents based on their specialized capabilities is a direct improvement over traditional static task allocation methods, leading to improved task outcomes.

[0057] Various example implementations are described herein with respect to the accompanying Figures.

[0058] Figure 1 illustrates a high-level overview of an embedding generation process. The process begins with task data 102, which can consist of various types of data descriptive of tasks. Examples of task data 102 can include academic benchmarks or real-time operational data from production systems.

[0059] The task data 102 is input into a rollout generation system 106. The rollout generation system 106 can be configured to interact with a number of different computer- implemented agents 104. Each computer-implemented agent 104 can be a software system capable of performing tasks that require complex reasoning. An example computer- implemented agent 104 can include one or more machine-learned models (e.g., sequence processing models) trained to process natural language queries and / or other input data to generate outputs.

[0060] The rollout generation system 106 utilizes the task data 102 and the capabilities of the computer-implemented agent 104 to generate execution rollout(s) 108 for the agent. Execution rollout(s) 108 can include detailed logs of the actions taken by the computer-implemented agent 104 in response to the task data 102. These logs can detail the sequence of actions and the intermediate results produced.

[0061] The execution rollout(s) 108 are then processed by an embedding generation system 110. The embedding generation system 110 can be designed to analyze the execution rollout(s) 108 generated for various combinations of agents and task data to generate a numerical representation or embedding of each agent’s behavior and capabilities. This system can use various machine learning techniques, such as autoencoders or triplet loss algorithms, to produce the embeddings.

[0062] Finally, the output from the embedding generation system 110 is an agent embedding 112. Agent embedding 112 can be a vector that represents the behavior and capabilities of the computer-implemented agent 104 based on the execution rollout(s) for agent 108. Agent embedding 112 can be stored for later use in matching agents with tasks or for other analytical purposes.

[0063] Figure 2 presents an example of an autoencoder structure used in the generation of agent embeddings. The process begins with execution rollout(s) for agent 202, which can include sequences of actions and results produced by an agent while performing tasks.

[0064] The execution rollout(s) for agent 202 are input into an embedding generation system 204. Within the embedding generation system 204, an encoder 206 processes the execution rollout(s) for agent 202. The encoder 206 can transform the input data into a compressed numerical form, known as an agent embedding 210. Examples of techniques used by the encoder 206 could include neural networks or other machine learning algorithms.

[0065] The agent embedding 210, which represents a condensed version of the agent’s behavior, is then processed by a decoder 208. The decoder 208 attempts toreconstruct the original execution rollout(s) for agent 202 from the agent embedding 210. The output of this process is reconstructed rollout(s) 212, which are an approximation of the original execution rollout(s) for agent 202.

[0066] The quality of the reconstruction by the decoder 208 is evaluated using an autoencoder loss 214. The autoencoder loss 214 measures the difference between the original execution rollout(s) for agent 202 and the reconstructed rollout(s) 212. This loss is used to adjust the parameters of the encoder 206 and decoder 208 to improve the accuracy of the agent embedding 210 and the reconstruction process.

[0067] Figure 3 illustrates the use of a triplet training approach in the generation of agent embeddings. The diagram shows three sets of execution rollouts being processed: first execution rollout(s) for agent 1 302. second execution rollout(s) for agent 1 304, and execution rollout(s) for agent 2 306.

[0068] Each set of execution rollouts is input into an embedding generation system 310. The embedding generation system 310 processes these inputs to generate embeddings for the agents. The first execution rollout(s) for agent 1 302 results in the first agent embedding for agent 1 322. Similarly, the second execution rollout(s) for agent 1 304 leads to the second agent embedding for agent 1 324. The execution rollout(s) for agent 2 306 is used to generate the agent embedding for agent 2 326.

[0069] These embeddings are then utilized in a triplet loss function 330. The triplet loss function 330 compares the first agent embedding for agent 1 322, the second agent embedding for agent 1 324, and the agent embedding for agent 2 326. The purpose of this function is to optimize the embeddings by minimizing the distance between embeddings of the same agent (agent 1) and maximizing the distance between embeddings of different agents (agent 1 and agent 2). The triplet loss function 330 can be used to update or otherwise train the embedding generation system 310. This training approach helps to refine the embeddings to better represent the distinct behaviors and capabilities of each agent.

[0070] Figure 4 depicts the use of a contrastive loss approach in the generation of agent embeddings. The process starts with the first execution rollout(s) for agent 1 402, which are input into two different systems.

[0071] The first input from the first execution rollout(s) for agent 1 402 goes directly into an embedding generation system 410. This system processes the input to produce the first agent embedding for agent 1 422. The first agent embedding for agent 1 422 represents a numerical vector encapsulating the behavior and capabilities of agent 1 based on the initial execution rollouts.

[0072] Simultaneously, the same first execution rollout(s) for agent 1 402 are also input into an augmentation system 408. The augmentation system 408 modifies or transforms the execution rollouts to create a varied version of the original data. This modified data is then fed into another instance of the embedding generation system 410, which generates the second agent embedding for agent 1 424. The second agent embedding for agent 1 424 represents an alternative numerical vector for the same agent, derived from the augmented execution rollouts. In some implementations, the rollout(s) 402 shown in the upper row can be augmented as well.

[0073] Both embeddings, the first agent embedding for agent 1 422 and the second agent embedding for agent 1 424, are then used in a contrastive loss function 430. The contrastive loss function 430 evaluates the similarity between the two embeddings, aiming to minimize the distance between embeddings derived from similar or augmented data of the same agent. This approach helps refine the embeddings to ensure they represent the agent’s capabilities even when faced with varied input data scenarios.

[0074] Figure 5 illustrates the process of generating intermediate embeddings from portions of execution rollouts, followed by their aggregation into a final agent embedding. The process begins with execution rollout(s) for agent 502, which are segmented into multiple parts, each represented as state, action, (result) 504-1, state, action, (result) 504-2, through state, action, (result) 504-n. These segments capture specific instances of the agent’s behavior during task execution.

[0075] Each segment, such as state, action, (result) 504-1 , is processed by an embedding generation system 510. This system is responsible for transforming the data from each segment into an intermediate embedding. As a result, state, action, (result) 504-1 is converted into intermediate embedding 512-1, state, action, (result) 504-2 into intermediate embedding 512-2, and similarly for other segments up to intermediate embedding 512-n.

[0076] Following the generation of intermediate embeddings, an embedding aggregation system 514 is used. This system aggregates all the intermediate embeddings, such as intermediate embedding 512-1, intermediate embedding 512-2, and intermediate embedding 512-n, to produce a single consolidated agent embedding 516. For example, embeddings can be averaged, clustered, etc.

[0077] The agent embedding 516 represents a comprehensive numerical vector that encapsulates the overall behavior and capabilities of agent 502 based on the detailed analysis of each action-result sequence in the execution rollouts.

[0078] Figure 6 depicts an example technique for generating task embeddings alongside agent embeddings. The diagram features two primary inputs processed by the same type of system but for different purposes.

[0079] The first input, execution rollout(s) for task 1 602, consists of data representing the actions and results of an agent performing task 1. This data is fed into an embedding generation system 610, which processes the execution rollouts to generate agent embedding(s) 622. Agent embedding(s) 622 serve as a numerical representation of the agent’s behavior and effectiveness in handling task 1.

[0080] Simultaneously, task data for task 1 604, which can include descriptions, parameters, or other relevant data about task 1, is also input into another instance of the embedding generation system 610. This system processes the task data to produce task embedding(s) 624. Task embedding(s) 624 encapsulate the characteristics and requirements of task 1 in a numerical form.

[0081] Both types of embeddings, agent embedding(s) 622 and task embedding(s) 624, are then utilized in a contrastive loss function 630. The contrastive loss function 630 aims to minimize the discrepancy between these embeddings. This mechanism helps in refining both agent and task embeddings for better alignment and performance prediction. In particular, by performing the illustrated approach over a large number of examples of tasks and rollouts, the embedding generation system can leam to generate embeddings within a shared embedding space that captures the relationships between tasks, between agents, and between tasks and agents.

[0082] Figure 7 illustrates an example technique for searching and / or routing a task query to one or more agents. The process starts with task data for new task 704, which includes specific details or requirements of a new task that needs to be addressed.

[0083] This task data for new task 704 is input into an embedding generation system 710. The embedding generation system 710 processes the input data to generate a task embedding 724. Task embedding 724 is a numerical representation that encapsulates the characteristics and requirements of the new task.

[0084] Simultaneously, an agent search system 702 maintains a database or repository of agent embeddings 722. These embeddings represent the capabilities and behaviors of various agents, previously generated and stored.

[0085] The task embedding 724 and agent embeddings 722 are then input into an embedding comparison system 740. The embedding comparison system 740 compares thetask embedding 724 with each of the agent embeddings 722 to determine which agent(s) are best suited to handle the new task based on their capabilities and the task requirements.

[0086] Based on the results of this comparison, one or more agents are selected, represented by selected agent(s) 750. These agents are deemed most appropriate for executing the new task as determined by the similarity or match between the task embedding 724 and the agent embeddings 722. This selection process helps in efficiently routing the task query to the most suitable agent(s) available.

[0087] Figure 8 depicts a flowchart of a method 800 for training one or more machine-learned models according to aspects of the present disclosure. For instance, an example machine-learned model can include an embedding generation model.

[0088] One or more portion(s) of example method 800 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 800 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 800 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 ordinary7skill 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 / tenns 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 800 can be performed additionally, or alternatively, by other systems.

[0089] At 802, example method 800 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 800 as a “training” instance, it is to be understood that runtime inferences can form training instances when a model is trained 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.

[0090] At 804, example method 800 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can bedirectly 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.

[0091] At 806, example method 800 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).

[0092] At 808, example method 800 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 800 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0093] In some implementations, example method 800 can be implemented for training a machine-learned model from an initialized state to a fully trained state (e.g., when the model exhibits a desired performance profde, such as based on accuracy, precision, recall, etc.).

[0094] In some implementations, example method 800 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 800 can be implemented for pre-training a machine-learned model. Pre-training caninclude, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety of tasks / data types.

[0095] In some implementations, example method 800 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 800 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.

[0096] In some implementations, example method 800 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.

[0097] An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use.

[0098] Figure 9 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.

[0099] 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 include decision tree based models, support vector machines, hidden Markov models, Bayesian netw orks, linear regression models, k-means clustering models, etc.

[0100] 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.

[0101] 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 networks, or other forms of neural networks. 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.

[0102] 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.

[0103] 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).

[0104] Machine-learned model(s) 1 can employ a mixture-of-experts structure. See, e.g., Zhou et al., Mixture-of-Experts with Expert Choice Routing, ARXlV:2202.09368v2 (Oct. 14, 2022). For example, different portions of a model can leam (explicitly or implicitly) different expertise areas, w ith 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 netw ork 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 w eights can form an '‘expert’’ that is selected by the router. On each forw ard pass, only a subset of the total model w eights 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.

[0105] 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 ty pe(s) or of different types of data as compared to input(s) 2. Output(s) 3 can include one type or many different types of data.

[0106] Example data ty pes 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.

[0107] 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.

[0108] An example input 2 can include one or multiple data t pes, such as the example data ty pes 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 or different 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.

[0109] Figure 10 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-M, 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.

[0110] 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, GOOG E, 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, ARXlV:2301.11325vl (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.

[0111] 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 sequence processing 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”).

[0112] 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.

[0113] 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.

[0114] For example, elements 5-1, 5-2, . . . , 5-M can represent tokens obtained using a tokenizer. 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-M) 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 / D18-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image.

[0115] 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 10 can be the tokens or can be the embedded representations thereof.

[0116] 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.

[0117] 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 textual snippet, “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.”

[0118] 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, ARXlV: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 ormore 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).

[0119] 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 well 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.

[0120] 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 types in output sequence(s) 7.

[0121] 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.

[0122] 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.

[0123] 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).

[0124] 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.

[0125] Figure 11 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.

[0126] 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.

[0127] 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 inthe 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.

[0128] 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.

[0129] 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 a learned within a continuous embedding space.

[0130] 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).

[0131] 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 datatype data-to-sequence model can subdivide an input of that arbitrary datatype and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7, 8-8, 8-9, etc.).

[0132] 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.

[0133] Figure 12 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.

[0134] 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 across various 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.

[0135] 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.

[0136] 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.

[0137] 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 an 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).

[0138] 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.

[0139] 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 17-2 can 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.

[0140] 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.

[0141] 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-shotprompts (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.

[0142] 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.

[0143] 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).

[0144] 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 one or more training iterations. Workbench 15 can implement prompt engineering tools in development model 16.

[0145] 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 a 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.

[0146] 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 with additional 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.

[0147] 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 wade variety of machine-learned models. Example training techniques can correspond to the example training method 800 described above.

[0148] 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 ofa 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.

[0149] 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”).

[0150] 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 prompts that 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.

[0151] 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 instruction that initiate API calls to send or obtain data via external systems.

[0152] 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.

[0153] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. Forinstance, 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.

[0154] 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.

[0155] Figure 13 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 flowcan 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. FIG. 13 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. FIG. 13 is described with reference to elements / terms described withrespect to other systems and figures for exemplar}' 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.

[0156] 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.

[0157] 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).

[0158] 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 as satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0159] 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-tuned model 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.

[0160] 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 computationaloptimization 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.

[0161] Figure 14 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.

[0162] 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.

[0163] 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. Model host 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.

[0164] 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.

[0165] 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.

[0166] In some implementations, model host 31 can operate on a 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 a 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.

[0167] 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 on 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 cached intermediate 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.

[0168] 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 asingle 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.

[0169] 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.

[0170] 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.

[0171] 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 the same 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.

[0172] 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.

[0173] 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.

[0174] 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) 1 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.

[0175] 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 is a 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 imageprocessing 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.

[0176] 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).

[0177] 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 speech recognition 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.

[0178] 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.

[0179] 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.

[0180] 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 an example, 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) I 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.

[0181] 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 orstorage (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.

[0182] 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.

[0183] 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.

[0184] 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 satisfying the 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 processand 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.

[0185] 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.

[0186] 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 imagery related to the context. For instance, machine-learned model(s) I 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).

[0187] 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 probability determined based on the context).

[0188] 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 probability determined based on the context).

[0189] Figure 15 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 perform 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 present disclosure (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 machine-learned 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.).

[0190] Network 49 can be any t pe of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof andcan include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of wired or wireless connection, using a wide 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 15 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0191] 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 with computing device 50).

[0192] 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 or multiple features described herein. The operations can implement example methods and techniques described herein.

[0193] 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.

[0194] 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.

[0195] 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.

[0196] 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.

[0197] Sen' er 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.

[0198] 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-serverrelationship 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.

[0199] 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 memory devices, 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 operations can 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.

[0200] 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 andtechniques 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).

[0201] Figure 15illustrates 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).

[0202] Figure 16 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 16, 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.

[0203] Figure 17 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).

[0204] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in Figure 17, 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.

[0205] 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 17, 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).

[0206] The technology discussed herein makes reference to servers, databases, software applications, and other computer-based systems, as well as actions taken and information 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.

[0207] 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 precludeinclusion 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.

[0208] 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 way 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.”

[0209] 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 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.

[0210] 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 method for generating a learned embedding space for computer- implemented agents, the method comprising: obtaining, by a computing system comprising one or more computing devices, a task dataset comprising data descriptive of a plurality of tasks; generating, by the computing system, a plurality of execution rollouts for a plurality of computer-implemented agents based on the task dataset, wherein each of the plurality of execution rollouts comprises a series of one or more outputs generated by one of the computer-implemented agents when performing one of the plurality of tasks; generating, by the computing system, a plurality of agent embeddings for the plurality of computer-implemented agents, wherein the agent embedding for each computer- implemented agent is learned based on one or more of the execution rollouts that are associated with the computer-implemented agent; and providing, by the computing system, the plurality of agent embeddings for the plurality of computer-implemented agents as an output.

2. The method of any preceding claim, wherein at least some of the plurality' of execution rollouts comprise a series of tuples comprising a state, an action, and a result.

3. The method of any preceding claim, wherein at least some of the plurality7of execution rollouts exhibit multi-step reasoning.

4. The method of any preceding claim, wherein at least some of the plurality of execution rollouts exhibit tool use.

5. The method of any preceding claim, wherein generating at least some of the plurality of execution rollouts comprises executing the computer-implemented agents in a sandbox computational environment.

6. The method of any preceding claim, wherein generating at least some of the plurality of execution rollouts comprises selecting one of the plurality of computer- implemented agents and one of the plurality of tasks at random.

7. The method of any preceding claim, wherein generating at least some of the plurality of execution rollouts comprises selecting one of the plurality of computer- implemented agents and one of the plurality of tasks based on metadata that indicates that the selected agent is configured for performing the selected task.

8. The method of any preceding claim, wherein generating, by the computing system, the plurality of agent embeddings for the plurality of computer-implemented agents comprises training an autoencoder model on the plurality of execution rollouts, wherein the autoencoder model comprises an encoder that learns to process an input execution rollout to generate an embedding and a decoder that learns to process the embedding to reconstruct the input execution rollout.

9. The method of any preceding claim, wherein generating, by the computing system, the plurality of agent embeddings for the plurality of computer-implemented agents comprises performing a triplet training algorithm to generate the plurality of agent embeddings, wherein the triplet training algonthm trains an embedding generation model to minimize a distance between embeddings generated for input execution rollouts associated with the same computer-implemented agent and to maximize a distance between embeddings generated for input execution rollouts associated with different computer-implemented agents.

10. The method of any preceding claim, wherein generating, by the computing system, the plurality' of agent embeddings for the plurality of computer-implemented agents comprises generating a plurality of intermediate embeddings for the plurality of execution rollouts and, for each computer-implemented agent, averaging the intermediate embeddings respectively associated with the execution rollouts that are performed by such computer- implemented agent to generate the agent embedding for such computer-implemented agent.

11. The method of any preceding claim, wherein the method further comprises generating a plurality of task embeddings respectively for the plurality of tasks.

12. The method of any preceding claim, further comprising: receiving an input query; generating a query embedding for the input query: andidentifying one or more of the computer-implemented agents as responsive to the input query based on a comparison of the query’ embedding to at least some of the plurality of agent embeddings.

13. The method of claim 12, wherein the input query is descriptive of a new task.

14. The method of claim 12 or 13, wherein identifying the one or more of the computer-implemented agents as responsive to the input query comprises directly comparing the query embedding to the at least some of the plurality of agent embeddings.

15. The method of claim 12 or 13, wherein identifying the one or more of the computer-implemented agents as responsive to the input query comprises evaluating a mapping from the query embedding to the learned embedding space containing the plurality of agent embeddings for the plurality of computer-implemented agents.

16. A computing system, comprising: one or more processors; and one or more non-transitory computer-readable media that collectively store: a plurality of agent embeddings respectively associated with a plurality of computer-implemented agents; and instructions that, when executed by the one or more processors, cause the computing system to perform operations, the operations comprising: receiving an input query; generating a query embedding for the input query; and identifying one or more of the computer-implemented agents as responsive to the input query based on a comparison of the query embedding to at least some of the plurality of agent embeddings.

17. The computing system of claim 16, yvherein the operations further comprise providing the input query for processing by at least one of the one or more computer- implemented agents that were identified as responsive to the input query'.

18. The computing system of claim 16 or 17, yvherein the input query comprises one or more of textual content, audio content, or image content, wherein the at least one of theone or more computer-implemented agents generates a response to the input query, and wherein the response comprises one or more of textual content, audio content, or image content.

19. The computing system of any of claims 16-18, wherein the operations further comprise processing the input query with the at least one of the one or more computer- implemented agents that were identified as responsive to the input query.

20. The computing system of claim 16, wherein the input query7comprises a description of a first agent, and wherein identifying the one or more computer-implemented agents as responsive to the input query comprises identifying the one or more computer- implemented agents as being similar to the first agent.

21. One or more non-transitory computer-readable media that collectively store a plurality of agent embeddings that have been generated as described in any of claims 1-15.

Images