Machine-learned model performing decoding operations based on a constraint status relating to a constraint associated with a query

Abstract

A computer-implemented method for a machine-learned model to respect a constraint provided in a query includes: receiving, by a computing system comprising one or more machine-learned models, a query which includes a constraint; and processing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

Description

MACHINE-LEARNED MODEL PERFORMING DECODING OPERATIONS BASED ON A CONSTRAINT STATUS RELATING TO A CONSTRAINT ASSOCIATED WITH A QUERYPRIORITY CLAIM

[0001] This application is based on and claims priority to United States Application Number 19 / 182.301 having a firing date of April 17, 2025 This application claims priority io and the benefit of United States Application Number 19 / 182,301 which is incorporated herein by reference in its entirety for ail purposes.FIELD

[0002] This disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the disclosure relates to machine-learned devices and systems comprising one or more machine-learned models receiving a constraint status after a first portion of an output is decoded and the one or more machine-learned models performing a subsequent decoding operation based on the constraint status to generate a second portion of the output.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 “learri’ to generate the desired outputs. The resulting model is often referred to as a machine-learned model.SUMMARY

[0004] Aspects and advantages of embodiments of the disclosure will be set forth in part in the following description, or can be learned from the description, or can be learned through practice of the embodiments.

[0005] Example aspects of the disclosure provide an example computer-implemented method. In some implementations, the example computer-implemented method can include: receiving, by a computing system comprising one or more machine-learned models, a querywhich includes a constraint; and processing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

[0006] In some implementations, the method includes encoding the constraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0007] In some implementations, the first portion includes one or more first soft tokens, and the second portion includes one or more second soft tokens.

[0008] In some implementations, the one or more machine-learned models perform a plurality7of decoding operations to generate a plurality7of tokens which form at least part of the output, and the one or more machine-learned models receive an update to the constraint status after each token is generated according to a respective decoding operation among the plurality of decoding operations.

[0009] In some implementations, the one or more machine-learned models perform a plurality of decoding operations to generate a plurality of tokens which form at least part of the output, and the one or more machine-learned models receive an update to the constraint status after a predetermined number of tokens are generated according to a predetermined number of decoding operations among the plurality of decoding operations.

[0010] In some implementations, the one or more soft tokens are provided to the one or more machine-learned models via a skip connection to one or more feed-forward layers of the one or more machine-learned models.

[0011] In some implementations, the constraint indicates a maximum number of tokens forming the output, and the one or more machine-learned models receive an update to the constraint status after one or more predetermined proportions of the maximum number of tokens are generated according to one or more decoding operations performed by the one or more machine-learned models among a plurality' of decoding operations performed by the one or more machine-learned models.

[0012] In some implementations, the constraint indicates a maximum time to generate the output, and the constraint status indicates a remaining time to generate the output.

[0013] In some implementations, the one or more machine-learned models perform a plurality of decoding operations to generate a plurality of tokens which form at least part of the output, and the one or more machine-learned models receive an update to the constraint status indicating the remaining time to generate the output after each token is generated according to a respective decoding operation among the plurality of decoding operations.

[0014] In some implementations, the one or more machine-learned models receive an update to the constraint status after one or more predetermined proportions of the maximum time to generate the output according to one or more decoding operations performed by the one or more machine-learned models among a plurality of decoding operations performed by the one or more machine-learned models.

[0015] In some implementations, the one or more machine-learned models receive an update to the constraint status after one or more predetermined intervals of time elapse during decoding operations performed by the one or more machine-learned models among a plurality of decoding operations performed by the one or more machine-learned models.

[0016] In some implementations, the constraint indicates a maximum time to generate the output, and the constraint status indicates a remaining time to generate the output and an elapsed time that the one or more machine-learned models has spent performing decoding operations in generating the output, and the method further comprises: encoding the constraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0017] In some implementations, the constraint indicates a maximum number of tokens forming the output, and the constraint status indicates a remaining number of tokens before the maximum number of tokens is reached and a number of tokens the one or more machine-learned models has currently output by performing decoding operations in generating the output, and the method further comprises: encoding the constraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0018] Example aspects of the disclosure provide an example computer-implemented method for training one or more machine-learned models. In some implementations, the example computer-implemented method can include: obtaining, by a computing system comprisingone or more processors and one or more machine-learned models, a query and one or more constraints associated with the query; processing, by the one or more machine-learned models, the query to generate a model-predicted generated content; evaluating the model-predicted generated content based on one or more ground truth training examples and based on whether the model-predicted generated content satisfied the one or more constraints, to determine a loss function; and training, by the computing system, the one or more machine-learned models based on the loss function.

[0019] Example aspects of the disclosure provide an example computing system or computing device that includes one or more processors; and one or more non-transitory computer-readable media storing: one or more machine-learned models; and instructions that, when implemented, cause the one or more processors to perform operations, the operations comprising: receiving a query which includes a constraint; and processing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

[0020] In some implementations, the operations further comprise encoding the constraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0021] In some implementations, the first portion corresponds to a first token, and the second portion corresponds to a second token.

[0022] In some implementations, the one or more machine-learned models perform a plurality of decoding operations to generate a plurality' of tokens which form at least part of the output, and the one or more machine-learned models receive an update to the constraint status after each token is generated according to a respective decoding operation among the plurality of decoding operations.

[0023] In some implementations, the constraint indicates a maximum time to generate the output, and the constraint status indicates a remaining time to generate the output and an elapsed time that the one or more machine-learned models has spent performing decoding operations in generating the output, and the operations further comprise: encoding theconstraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0024] In some implementations, the constraint indicates a maximum number of tokens forming the output, and the constraint status indicates a remaining number of tokens before the maximum number of tokens is reached and a number of tokens the one or more machine-learned models has currently output by performing decoding operations in generating the output, and the operations further comprise: encoding the constraint status as one or more soft tokens; and providing the one or more soft tokens to the one or more machine-learned models.

[0025] The example computing system or computing device may be configured to perform any other aspects and operations of the computer-implemented methods as described herein.

[0026] Example aspects of the disclosure provide one or more example non-transitory computer-readable media storing one or more machine-learned models and instructions that are executable by one or more processors to cause a computing system to perform example operations. In some implementations, the example operations can include receiving a query which includes a constraint; and processing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

[0027] The non-transitory computer-readable medium may store additional instructions to execute other aspects and operations of the computing systems, computing devices, and computer-implemented methods as described herein.

[0028] Other example aspects of the disclosure are directed to other systems, methods, apparatuses, tangible non-transitory computer-readable media, and devices for performing functions described herein. These and other features, aspects, and advantages of various implementations will become better understood with reference to the following description and appended claims. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate implementations of the disclosure and. together with the description, help explain the related principles.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] FIG. 1 A is an example system, according to one or more example embodiments of the disclosure;

[0030] FIG. IB is an example block diagram of a computing system, according to one or more example embodiments of the disclosure;

[0031] FIGS. 2A-2B illustrate flow diagrams of example, non-limiting computer-implemented methods, according to one or more example embodiments of the disclosure;

[0032] FIG. 3 illustrates an example block diagram or architecture of an environment for generating content via one or more machine-learned models based on a query and a constraint associated with the query, according to one or more example embodiments of the disclosure;

[0033] FIG. 4 illustrates a flow diagram of an example, non-limiting computer-implemented method, according to one or more example embodiments of the disclosure;

[0034] FIGS. 5A and 5B illustrate example block diagrams or architectures of environments for training or implementing one or more machine-learned models, according to one or more example embodiments of the disclosure;

[0035] FIG. 6 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the disclosure;

[0036] FIG. 7 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 disclosure;

[0037] FIG. 8 is a block diagram of an example sequence processing model according to example implementations of aspects of the disclosure;

[0038] FIG. 9 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 disclosure;

[0039] FIG. 10 is a block diagram of an example model development platform according to example implementations of aspects of the disclosure;

[0040] FIG. 11 is a block diagram of an example training workflow for training a machine-learned model according to example implementations of aspects of the disclosure;

[0041] FIG. 12 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 disclosure;

[0042] FIG. 13 is a block diagram of an example networked computing system according to example implementations of aspects of the disclosure;

[0043] FIG. 14 is a block diagram of an example computing device according to example implementations of aspects of the disclosure; and

[0044] FIG. 15 is a block diagram of an example computing device according to example implementations of aspects of the disclosure.DETAILED DESCRIPTION

[0045] Reference now will be made to embodiments of the disclosure, one or more examples of which are illustrated in the drawings, wherein like reference characters denote like elements. Each example is provided by way of explanation of the disclosure and is not intended to limit the disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made to disclosure without departing from the scope or spirit of the disclosure. 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 disclosure covers such modifications and variations as come within the scope of the appended claims and their equivalents.

[0046] Machine-learned models can be implemented to perform various tasks including providing answers to queries, language translation, image recognition, natural language processing, etc. Machine-learned models (e.g., large language models, generative models, etc.), are implemented in various environments and applications, including Al assistants such as Gemini and ChatGPT. In some implementations, when the machine-learned model receives an input (a query) and processes the input to provide an output, the machine-learned model may be configured to decode the output one token at a time, iterating until a stopping condition is met, such as the machine-learned model outputting an end-of-sequence (EOS) token.

[0047] However, in some instances a user or computing system may want or need to constrain the output to be limited to a certain number of tokens, sentences, paragraphs, etc. Additionally, or alternatively, the user or computing system may want or need the machine-learned model to provide the output within a certain duration of time (e.g., a latency less than some threshold value).

[0048] For example, based on receiving a query (e.g., "What are the rules of soccer?" or "Who were the last 10 US presidents?"), the machine-learned model may provide an output that is concise with a few tokens, or generate a long and detailed output. In this regard, each token generated by the machine-learned model has an associated computing cost, as generating tokens requires computing power, time, etc. In some implementations, a machine-learned model provider that provides the machine-learned model for use by an entity may charge a fee for each token that is generated. Therefore, configuring the machine-learned model to limit or constrain the number of tokens generated for an output (e g., based on a user input or user-defined constraint) can conserve or save processing power, energy, and fees associated with generated the output, while also satisfying expectations of a user.

[0049] Each token generated by the machine-learned model also has an associated latency such that an output provided by the machine-learned model can take a certain duration of time. In some implementations, the machine-learned model may be configured to limit or constrain the duration of time taken to provide an output (e.g., based on a user input or user-defined constraint), which can also conserve or save processing power and energy, while also satisfying expectations of a user. Controlling latency during inference time for the machine-learned model can also be applied and can further take into account circumstances where the machine-learned model performs external tool and API calls.

[0050] Current methods exist for limiting the output of a machine-learned model. For example, in one method a parameter specifying the maximum number of decoding steps (or the maximum time allowed for decoding) may be configured for the machine-learned model, such that once the maximum number of decoding steps (or the maximum time) is reached, the model inference is cut off. However, while the constraint may be satisfied, this method is problematic since the machine-learned model is not aware of how many decoding steps (or time) it has left, and thus the output generated by the machine-learned model once the constraint is met is often incomplete and is not fully responsive to the user query.

[0051] In another method, the machine-learned model may be given a prompt to output at most a certain number of tokens. However, this method is often deficient because the machine-learned model may have difficulty keeping track of the number of tokens or may not be trained to respect the instruction exactly. Further, in prior methods machine-learned models are unable to respect latency requirements since the models do not have a sense of time.

[0052] In view of the above challenges, aspects of the disclosure are directed to computer-implemented methods and computing systems that can honor token and / or time constraints described in a prompt by making the machine-learned model aware of the number of decoding steps remaining and / or of the time remaining for providing an output. The systems and methods described herein can be applied to various machine-learned models including sequence models, large language models, generative machine-learned models, neural networks, etc.

[0053] In particular, an aspect of the disclosure is directed to a computing system including a machine-learned model which receives a query indicating an output constraint and, at each decoding step, the machine-learned model receives information indicating a decoding reserve or decoding margin (e.g., the remaining capacity before the constraint is met. such as the remaining step count and / or remaining time to provide the output). This allows the machine-learned model to plan how to use the decoding reserve (decoding margin or remaining capacity ) to end up with a complete output that is responsive to the query, while controlling costs and latency and satisfying user-defined constraints. For example, the computing system may include an encoder which is configured to encode information corresponding to a remaining step count or remaining tokens and / or remaining time, and to feed the information to the machine-learned model. In some implementations, the encoder can additionally, or alternatively, encode information corresponding to how many steps have been performed or tokens generated and / or an elapsed time that has passed by during the decoding process, and feeds the information to the machine-learned model.

[0054] In some implementations, the computing system may include a timer which tracks the duration of time that the machine-learned model has remaining to provide the output based on a latency constraint which can be included in the prompt to the machine-learned model. The time information can be provided to the encoder which may be configured to encode the information corresponding to the remaining time for providing the output and to feed the information to the machine-learned model.

[0055] In some implementations, the encoder is configured to encode information corresponding to the remaining step count and / or the remaining time at each decoding step or operation and to feed the information to the machine-learned model at each decoding step or operation. In some implementations, the encoder is configured to encode information corresponding to the remaining step count and / or the remaining time at predetermined decoding intervals or operations and to feed the information to the machine-learned model at the predetermined decoding intervals or operations (e.g., every three tokens, every fivetokens, every ' / i second, every second, etc ). In some implementations, the encoder is configured to encode information corresponding to the remaining step count and / or the remaining time at a predetermined proportion(s) of the decoding step or operation constraint and to feed the information to the machine-learned model at the predetermined proportion(s) of the decoding step or operation (e.g., at reaching 1 / 3 of the token constraint, at reaching 1 / 5 of the token constraint, at reaching 1 / 3 of the time constraint, at reachingx / i of the time constraint, etc.).

[0056] In some implementations, an aspect of the disclosure is directed to training the machine-learned model to respect token and / or time constraints included in a prompt. The machine-learned model can be trained to provide an output that satisfies expectations of the user and satisfies constraint conditions included in a query, where the output is generated by the machine-learned model based on inputting encoded information regarding a constraint status during the decoding process.

[0057] For example, the machine-learned model can be trained by providing different inputs (different queries) and the output of the machine-learned model can be scored using a reward model. For example, the reward model may be configured to reward or penalize the machine-learned model based on whether the output corresponds to the expected output (e g., ground truth output) and satisfies the constraint.

[0058] For example, if the machine-learned model does not output an end-of-sequence (EOS) token before the constraint was reached, the computing system can penalize the machine-learned model. For example, the penalty may be proportional to the number of additional tokens in excess of the constraint or proportional to the time needed to generate the output in excess of the constraint, or the penalty may be non-linear (e.g., a quadratic error). If the machine-learned model does output the EOS token before the constraint was reached, the computing system and / or a user may analyze the output sequence to see whether the output is both coherent and completely answers the query. Evaluating the output can be done automatically (e.g., via an autorater) or by having the user provide feedback (e.g., via RLHF). If the output was not coherent or did not fully answer the query, the reward model or user can provide a penalty to the machine-learned model.

[0059] The disclosed methods and systems yield several technical benefits. One significant benefit is enhanced technical control over the computer-implemented generative process executed by the machine-learned model during decoding. By incorporating real-time feedback regarding the constraint status (e.g., remaining computational budget like decoding steps / tokens or time), the system dynamically guides the subsequent steps of the generationalgorithm. This allows the model to actively manage its generation strategy, resulting in the production of a more complete and semantically coherent output that adheres to the specified constraint boundaries. Compared to approaches where constraints are applied externally or post-hoc, this method significantly improves the functional quality of the output under constraints. Mechanisms such as feeding a soft token representing the constraint status as an input to the machine-learned model facilitate this improved awareness and control within the model's technical architecture.

[0060] As used herein, a "soft token" may include an embedding vector configured for processing by sequence-processing layers of a machine-learned model, often as part of an input sequence representation, wherein the embedding vector does not correspond to a fixed entry in a predefined model vocabulary7(e.g., a vocabulary7used for text tokenization).Consequently, the values of the embedding vector may be determined based on optimization during a training or tuning process (e.g., prompt tuning), or may be dynamically computed at inference time based on runtime state or other conditioning information. Such an embedding vector is often processed directly by the model's layers, distinct from processes involving quantization against a finite codebook to yield a discrete index. A soft token can represent, for example, task-specific instructions, control signals, learned prompt representations, continuous data representations derived from non-textual modalities (e.g., images), or runtime state information such as the constraint status described herein, thereby enabling the injection of nuanced conditioning signals into the model. The soft token embedding vector ty pically possesses a dimensionality compatible with the model's standard token embeddings, facilitating its integration into the model's processing architecture.

[0061] This improved control directly contributes to a more efficient allocation and use of computational resources. Because the model is aware of impending constraint boundaries during generation, the system avoids computation associated with generating data segments (e.g., tokens, intermediate representations) or consuming processing time that would ultimately exceed the specified limit. This results in a reduction of unnecessary processing cycles (e.g., on CPUs, GPUs, TPUs), memory bandwidth usage, and potentially energy7consumption, leading to improved operational efficiency of the computing system performing the generation task.

[0062] Furthermore, the capability to incorporate and react to time-based constraints, providing the model with information on elapsed or remaining time, affords direct influence over the inference latency of the model's response generation. This technical capabilityfacilitates the deployment of complex generative models in technical systems operating under specific latency requirements, such as certain interactive control systems or time-sensitive data processing pipelines, where predictable response times are advantageous.

[0063] Consequently, the described methods and systems contribute to the overall reliability and predictability of the technical system implementing the machine-learned model.Downstream technical processes or end-users benefit from a higher likelihood of receiving a usable output that conforms to specified operational parameters (e.g., data size limits, response time limits). This enhanced functional reliability can further lead to downstream efficiencies in resource usage (e.g., network bandwidth, processing resources) by reducing the need for repeated requests or additional corrective processing that might arise from incomplete or non-compliant outputs.

[0064] Referring now to the drawings, FIG. 1 is an example system according to one or more example embodiments of the disclosure. In FIG. 1, the system 1100 includes a computing device 100, an external computing device 200, a server computing system 300. and external content 500, which may be in communication with one another over a network 400. For example, the computing device 100 and the external computing device 200 can serve as a local device or client device, and can include any of a personal computer, a smartphone, a tablet computer, a laptop, a global positioning service device, a smartwatch, and the like. The network 400 may include any type of communications network including a wired or wireless network, or a combination thereof. The network 400 may include a local area network (LAN), wireless local area network (WLAN), wide area network (WAN), personal area network (PAN), virtual private network (VPN), or the like. For example, wireless communication between elements of the example embodiments may be performed via a wireless LAN, Wi-Fi, Bluetooth, ZigBee, Wi-Fi direct (WFD), ultra wideband (UWB), infrared data association (IrDA), Bluetooth low energy (BLE), near field communication (NFC), a radio frequency (RF) signal, and the like. For example, wired communication between elements of the example embodiments may be performed via a pair cable, a coaxial cable, an optical fiber cable, an Ethernet cable, and the like. Communication over the network 400 can use a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), and / or protection schemes (e.g., VPN, secure HTTP, SSL).

[0065] As will be explained in more detail below, in some implementations the computing device 100 and / or server computing system 300 may form part of an application systemwhich may be configured to implement a machine-learned model that is trained to respect one or more constraints included in a query.

[0066] In some example embodiments, the server computing system 300 may obtain data from one or more of a content data store 360 and a machine-learned model data store 370, to implement various operations and aspects of the application systems as disclosed herein. The content data store 360 and machine-learned model data store 370 may be integrally provided with the server computing system 300 (e.g., as part of the one or more memory devices 320 of the server computing system 300) or may be separately (e.g.. remotely) provided. Further, content data store 360 and machine-learned model data store 370 can be combined as a single data store (database) or may include a plurality of respective data stores. Data stored in one data store may overlap with some data stored in another data store. In some implementations, one data store may reference data that is stored in another data store.

[0067] In some implementations, the content data store 360 can store data associated with content. For example, the content can include text, images, videos, etc. In some implementations, the information stored in the content data store 360 can be associated with and / or stored according to a particular user or a plurality of users, according to a particular content category, content genre, content context, time, location, content ty pe, content environment, etc. In some implementations, the information stored in the content data store 360 can be associated with and / or stored according to a particular entity that is associated with the content (e.g., an entity that requests the content to be generated, an entity that is to receive the content, an entity' that appears in the content, etc.). For example, machine-learned models described herein can reference or retrieve content from the content data store 360 when generating an output or response to a query.

[0068] Machine-learned model data store 370 can store machine-learned models which can be retrieved and implemented by the server computing system 300 for generating distilled or fine-tuned machine-learned models (e.g., distilled or fine-tuned generative machine-learned models) that, in some implementations, can also be provided to the computing device 100. Machine-learned model data store 370 can also store distilled or fine-tuned machine-learned models (e.g.. distilled or fine-tuned generative machine-learned models) which can be retrieved and implemented by the computing device 100. In some implementations, the computing device 100 can retrieve and implement machine-learned models which are large parameter models that have not been fine-tuned or distilled. The machine-learned models (including large parameter models and distilled or fine-tuned models) stored at the machine-learned model data store 370 can include generative machine-learned models respectively associated with different types of applications, types of items, etc., that may be implemented across a variety of domains (e.g., navigation, mapping, healthcare, gaming, engineering / science, entertainment, travel, retail, etc.). The machine-learned models may include large language models and general, multimodal models. The machine-learned models may include text-to-text large language models, text-to-image large language models, etc. The machine-learned models may include language models which have been trained using reinforcement learning from human feedback. The machine-learned models may include generative artificial intelligence (Al) models which may implement generative adversarial networks (GANs), transformers, variational autoencoders (VAEs), neural radiance fields (NeRFs). and the like.

[0069] External content 500 can be any form of external content including news articles, webpages, image files, video files, audio files, written descriptions, ratings, game content, social media content, photographs, commercial offers, transportation methods, weather conditions, sensor data obtained by various sensors, or other suitable external content. The computing device 100, external computing device 200, and server computing system 300 can access external content 500 over network 400. External content 500 can be searched by computing device 100, external computing device 200, and server computing system 300 according to known searching methods and search results can be ranked according to relevance, popularity, or other suitable attributes, including location-specific filtering or promotion.

[0070] FIG. IB is an example block diagram of a computing system, according to one or more example embodiments of the disclosure. Referring now to FIG. IB, example block diagrams of a system 1200 including a computing device 100 and server computing system 300 according to one or more example embodiments of the disclosure will now be described. Although computing device 100 is represented in FIG. IB, features of the computing device 100 described herein are also applicable to the external computing device 200.

[0071] The computing device 100 may include one or more processors 110, one or more memory devices 120, a content generation system 130, a position determination device 140, an input device 150, a display device 160, and an output device 170. The server computing system 300 may include one or more processors 310, one or more memory devices 320, and a content generation system 330.

[0072] For example, the one or more processors 110, 310 can be any suitable processing device that can be included in a computing device 100 or server computing system 300. For example, the one or more processors 110, 310 may include one or more of a processor, processor cores, a controller and an arithmetic logic unit, a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an image processor, a microcomputer, a field programmable array, a programmable logic unit, an applicationspecific integrated circuit (ASIC), a microprocessor, a microcontroller, etc., and combinations thereof, including any other device capable of responding to and executing instructions in a defined manner. The one or more processors 110, 310 can be a single processor or a plurality of processors that are operatively connected, for example in parallel.

[0073] The one or more memory' devices 120, 320 can include one or more non-transitory computer-readable storage mediums, including a Read Only Memory (ROM), Programmable Read Only Memory (PROM), Erasable Programmable Read Only Memory (EPROM), and flash memory, a USB drive, a volatile memory device including a Random Access Memory (RAM), a hard disk, floppy disks, a Blu-ray disk, or optical media such as CD ROM discs and DVDs, and combinations thereof. However, examples of the one or more memory¬ devices 120, 320 are not limited to the above description, and the one or more memory devices 120, 320 may be realized by other various devices and structures as would be understood by those skilled in the art.

[0074] In some example embodiments, the computing device 100 includes a content generation system 130. For example, the content generation system 130 may include the content generation application 132 and one or more machine-learned models 134. The computing device 100 can include various other applications including search applications, gaming applications, document applications, text messaging applications, email applications, dictation applications, virtual keyboard applications, browser applications, map applications, social media applications, etc. Aspects of the content generation application 132 can be implemented in other applications, for example, for generating content for an email application, document application, social media application, etc. In some implementations, the content generation application 132 may be part of another application (e.g., a search application, a generative application, health application, gaming application, etc.) or may be a standalone application. The content generation application 132 may be configured to be dynamically interactive according to various user inputs. Example implementations of the content generation application 132 are described herein, however the disclosure is not limitedto these examples as various modifications may be made to the embodiments described herein.

[0075] The one or more memory devices 120 can include data 122 and instructions 124 that can be retrieved, manipulated, created, or stored by the one or more processors 110. In some example embodiments, such data can be accessed and used as input to implement content generation application 132 which may be implemented together with the one or more machine-learned models 134, and to execute the instructions to perform operations including receiving, by the one or more machine-learned models 134. a query which includes a constraint; and processing, by the one or more machine-learned models 134, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models 134, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models 134, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models 134, a second decoding operation based on the constraint status, to generate a second portion of the output, as described according to examples of the disclosure.

[0076] For example, the one or more memory devices 320 can also include data 322 and instructions 324 that can be retrieved, manipulated, created, or stored by the one or more processors 310. In some example embodiments, such data can be accessed and used as input to implement content generation application 332 which may be implemented together with the one or more machine-learned models 334, and to execute the instructions to perform operations including receiving, by the one or more machine-learned models 334, a query which includes a constraint; and processing, by the one or more machine-learned models 334, the query to generate an output which satisfies the constraint, wherein the processing includes: performing, by the one or more machine-learned models 334, a first decoding operation to generate a first portion of the output, receiving, by the one or more machine-learned models 334, a constraint status after the first portion of the output is decoded, and performing, by the one or more machine-learned models 334, a second decoding operation based on the constraint status, to generate a second portion of the output, as described according to examples of the disclosure.

[0077] According to examples of the disclosure, the content generation application 132 may be executed by the computing device 100 to generate content based on a query received from a user and / or another computing system. In some implementations, the query’ can include aconstraint with respect to the output (content to be generated). In some implementations, the constraint can be defined or indicated by the user (e.g., “I need a 1000 word essay on the inventions of Benjamin Franklin”). In some implementations, the constraint can be defined or indicated by the computing system and included as part of the query or prompt (e.g., limiting the output to a certain length, limiting the time to generate the output to a certain timeframe, combinations thereof, etc.).

[0078] In some examples, one or more aspects of the content generation application 132 may be implemented by the content generation application 332 of the server computing system 300 which may be remotely located, to generate content. In some examples, one or more aspects of the content generation application 332 may be implemented by the content generation application 132 of the computing device 100, to generate content.

[0079] In some example embodiments, the computing device 100 includes aposition determination device 140. Position determination device 140 can determine a current geographic location of the computing device 100 and communicate the geographic location to the server computing system 300 over network 400. The position determination device 140 can be any device or circuitry for analyzing the position of the computing device 100. For example, the position determination device 140 can determine actual or relative position by using a satellite navigation positioning system (e.g. a GPS system, a Galileo positioning system, the GLObal Navigation satellite system (GLONASS), the BeiDou Satellite Navigation and Positioning system), an inertial navigation system, a dead reckoning system, based on an IP address, by using triangulation and / or proximity to cellular towers or WiFi hotspots, and / or other suitable techniques for determining aposition of the computing device 100. For example, in some implementations the content generation application 132 and / or one or more machine-learned models 134 may be configured to utilize position information determined by the position determination device 140 to generate output responsive to a query or prompt (e.g., an answer to a question, code, etc.).

[0080] The computing device 100 may include an input device 150 configured to receive an input from a user and may include, for example, one or more of a keyboard (e.g., a physical keyboard, virtual keyboard, etc.), a mouse, a joystick, a button, a switch, an electronic pen or stylus, a gesture recognition sensor (e.g., to recognize gestures of a user including movements of a body part), an input sound device or speech recognition sensor (e.g., a microphone to receive a voice input such as a voice command or a voice query), a track ball, a remote controller, a portable (e.g.. a cellular or smart) phone, a tablet PC, a pedal or footswitch, avirtual-reality device, and so on. The input device 150 may also be embodied by a touch-sensitive display having a touchscreen capability, for example. For example, the input device 150 may be configured to receive an input from a user associated with the input device 150 for executing the content generation application 132, for providing an input prompt to the content generation application 132 and / or one or more machine-learned models 134, for providing feedback to the content generation application 132 and / or one or more machine-learned models 134, etc.

[0081] The computing device 100 may include a display device 160 which displays information viewable by the user (e.g., a user interface screen). For example, the display device 160 may be anon-touch sensitive display or a touch-sensitive display. The display device 160 may include a liquid crystal display (LCD), a light emitting diode (LED) display, an organic light emitting diode (OLED) display, active matrix organic light emitting diode (AMOLED), flexible display, 3D display, a plasma display panel (PDP), a cathode ray tube (CRT) display, and the like, for example. However, the disclosure is not limited to these example displays and may include other types of displays. The display device 160 can be used by the content generation system 130 provided at the computing device 100 to display information for generating content via the one or more machine-learned models 134. The display device 160 can be configured to provide, for presentation to a user, one or more user interface screens having user interface elements which are selectable by the user for generating content according to an input, for providing feedback or instructions (guidance) to a user regarding operations for generating content via the one or more machine-learned models 134, etc.

[0082] The computing device 100 may include an output device 170 to provide an output to the user and may include, for example, one or more of an audio device (e.g., one or more speakers), a haptic device to provide haptic feedback to a user (e.g., a vibration device), a light source (e.g., one or more light sources such as LEDs which provide visual feedback to a user), a thermal feedback system, and the like. For example, the output device 170 may provide information relating to operations for controlling or selecting functions of the content generation application 132, including an output confirming selections made by a user, an output relating to generating content according to the one or more machine-learned models 134, for providing feedback or instructions (guidance) to a user regarding operations for generating content or training or refining the one or more machine-learned models 134, etc.

[0083] The computing device 100 may include fewer or more components than that shown in FIG. IB. For example, the computing device 100 can include one or more capture devices configured to capture media content (e.g., images, video, sound, etc.), one or more sensors configured to capture various data (e.g., motion data, environmental data, biometric data, etc.). For example, in some implementations, content which is captured by the one or more capture devices and / or the one or more sensors may be provided as an input to the content generation system 130 (e.g., the content generation application 132) to indicate a status of a user, a location, an environment, and / or of the computing device 100.

[0084] In accordance with example embodiments of the disclosure, the server computing system 300 can be a remote device that is remotely located from the computing device 100. The server computing system 300 can include one or more processors 310 and one or more memory devices 320 as described herein. The server computing system 300 may also include a content generation system 330 which is similar to the content generation system 130 described herein.

[0085] For example, the content generation system 330 may include the content generation application 332 which performs functions similar to those described herein with respect to content generation application 132 and the one or more machine-learned models 134, and therefore will not be repeated for the sake of brevity. In some implementations, the content generation application 332 may be part of another application (e.g., a search application, health application, gaming application, content generation application, etc.) or may be a standalone application.

[0086] Examples of the disclosure are directed to computer implemented methods for application systems, for example application systems including content generation applications, which are configured to generate content according to example described herein.

[0087] The flow diagrams of FIG. 2A-2B illustrate example methods for generating content based on a query while respecting one or more given constraints. Although shown in a particular sequence or order, unless otherwise specified, the order of the processes can be modified. Thus, the illustrated embodiments should be understood only as examples, and the illustrated processes can be performed in a different order, and some processes can be performed in parallel. Additionally, one or more processes can be omitted in variousembodiments. Thus, not all processes are required in every embodiment. Other process flows are possible.

[0088] The operation of FIG. 2A will be explained with reference to FIG. 3. FIG. 3 illustrates an example block diagram or architecture of an environment for generating content via one or more machine-learned models based on a query and a constraint associated with the query', according to one or more example embodiments of the disclosure.

[0089] Referring to FIG. 2A, at operation 2110 the method 2100 includes a computing system (e.g., including content generation application 132 and / or content generation application 332) receiving or obtaining a query and an associated constraint. In some examples the query’ can include the constraint. In some examples, the constraint may be internally generated by the computing system and associated with the query. For example, the query' may be to write a 500 word essay on George Washington (with the constraint being a 500 word limitation to the essay), or to generate, within a particular amount of time, code to perform a particular task (with the constraint being the indicated time for generating the code). As described herein, the computing system may be embodied as computing device 100 and / or server computing system 300, for example. In the environment 3100 of FIG. 3, an input 3110 is provided to the content generation application 3120 (which can correspond to the content generation application 132 and / or content generation application 332). The input 3110 may comprise the query which can include the constraint. The input 3110 may be provided to the one or more machine-learned models 3122 (which can correspond to the one or more machine-learned models 134 and / or the one or more machine-learned models 334).

[0090] As operation 2120 the method 2100 includes the computing system processing, via the one or more machine-learned models, the query to generate an output which satisfies the constraint. For example, in FIG. 3 the one or more machine-learned models 3122 may be configured to process the input 3110 to output the generated content 3130 which satisfies the constraint included in the input 3110. Details regarding example operations for how the one or more machine-learned models 3122 process the input 3110 are described with respect to FIG. 2B and FIG. 3.

[0091] The operation of FIG. 2B will be explained w ith reference to FIG. 3. Referring to FIG. 2B, at operation 2210 the method 2200 includes the computing system (e.g., including content generation application 132 and / or content generation application 332) performing a first decoding operation to generate a first portion of the output. For example, referring toFIG. 3, in processing the input 3110 the one or more machine-learned models 3122 may be configured to generate a first portion (e.g., a first token) which forms part of the output which is eventually generated as the generated content 3130.

[0092] In some implementations, the one or more machine-learned models 3122 may be configured to receive the input 3110 (e.g., a query' or prompt including the constraint), and follow a structured process to generate a response token by token. That is, the one or more machine-learned models 3122 may be configured to receive the input 3110 as a string of text, tokenize the input, and convert the tokenized input into a numerical form using an embedding layer, for example, by mapping the tokens to a high-dimensional vector. The one or more machine-learned models 3122 may be configured to generate a start token and predict the first token which can form part of the response to the query. The one or more machine-learned models 3122 may be configured to utilize various methods for predicting the token, including a greedy decoding method which selects a token with the highest probability, a beam search method which keeps track of multiple sequences (beams) to select better overall sequences, a top-k sampling method which randomly samples from the top-k most probable tokens, combinations thereof, etc.

[0093] At operation 2220 the method 2200 includes the computing system receiving a constraint status after the first portion of the output is decoded. For example, in some implementations after generating the first portion (e.g., the first token), the content generation application 3120 may be configured to track or monitor a status of the constraint.

[0094] In some implementations, the token counter 3124 may be configured to count the number of tokens that have been generated thus far during the decoding process and provide an output to the encoder 3128 indicating the constraint status. In some implementations, the token counter 3124 may be configured to additionally, or alternatively, count the number of tokens remaining before a constraint is reached and provide an output to the encoder 3128 indicating the constraint status (e g., if the constraint indicates the output should be limited to 500 words and 10 words forming part of the response have been generated the token counter 3124 may be configured to provide an output that indicates 490 tokens are remaining and / or 10 tokens have been generated).

[0095] In some implementations, the timer 3126 may be configured to keep track of the time that has passed during the decoding process and provide an output to the encoder 3128 indicating the constraint status. In some implementations, the timer 3126 may be configuredto additionally, or alternatively, keep track of the time remaining before a constraint is reached and provide an output to the encoder 3128 indicating the constraint status (e.g., if the constraint indicates the output should be generated in 1 second and 50 ms have passed during the decoding process, the timer 3126 may be configured to provide an output that indicates 950 ms are remaining and / or that 50 ms have passed).

[0096] For example, the constraint status can also indicate a time spent for each decoding operation (e.g., to decode a token), an average time spent on each decoding operation (e.g., to decode a token), an average time that it takes to obtain information (e.g., via a call operation) from an external tool (e.g., via an API) for performing a decoding operation (e.g., 2 seconds), etc. Therefore, the one or more machine-learned models 3122 are given information during the decoding process that makes the one or machine-learned models 3122 aware of how much time is left and / or how many tokens remain to provide a final output that is complete and cohesive. In some implementations, the content generation application 3120 (the one or more machine-learned models 3122) may include a corrector network that is configured to update the encoding of the constraint status for more efficiently updating a stale embedding.

[0097] The encoder 3128 may be configured to provide the constraint status as an input to the one or more machine-learned models 3122 for a next or subsequent decoding operation. For example, the encoder 3128 may be configured to encode the constraint status as a soft token and provide the soft token to the one or more machine-learned models 3122. In some implementations, the constraint status information (e.g., the soft token) may be passed to the one or more machine-learned models 3122 as another token that attends to all other tokens or to a subset of the tokens, e.g. the most recent ones. In some implementations, the constraint status information (e.g., the soft token) may be passed or provided to the one or more machine-learned models 3122 with a skip connection to one or more feed-forward layers of the one or more machine-learned models 3122, thus reducing attention cost.

[0098] In some implementations, the encoder 3128 is configured to encode information corresponding to the remaining step or token count and / or the remaining time at each decoding operation and to feed the information to the one or more machine-learned models 3122 at each decoding operation. In some implementations, the encoder 3128 is configured to encode information corresponding to the remaining step or token count and / or the remaining time at predetermined decoding intervals or operations and to feed the information to the one or more machine-learned models 3122 at the predetermined decoding intervals or operations (e.g., every three tokens, every five tokens, every 'A second, every second, etc.).In some implementations, the encoder 3128 is configured to encode information corresponding to the remaining step or token count and / or the remaining time at one or more predetermined proportions of the constraint and to feed the information to the one or more machine-learned models 3122 at the one or more predetermined proportions of the constraint (e.g., at reaching 1 / 3 of the maximum token constraint, at reaching 1 / 5 of the maximum token constraint, at reaching 1 / 3 of the time constraint, at reaching ! of the time constraint, etc.).

[0099] In some implementations, the encoder 3128 is additionally, or alternatively, configured to encode information corresponding to the step or token count and / or the elapsed time at each decoding operation and to feed the information to the one or more machine-learned models 3122 at each decoding operation. In some implementations, the encoder 3128 is configured to encode information corresponding to the step or token count and / or the elapsed time at predetermined decoding intervals or operations and to feed the information to the one or more machine-learned models 3122 at the predetermined decoding intervals or operations (e.g., every three tokens, every five tokens, every Vi second, every second, etc.). In some implementations, the encoder 3128 is configured to encode information corresponding to the step or token count and / or the elapsed time at a predetermined proportion(s) of the constraint and to feed the information to the one or more machine-learned models 3122 at the predetermined proportion(s) of the constraint (e.g., at reaching 1 / 3 of the token constraint, at reaching 1 / 5 of the token constraint, at reaching 1 / 3 of the time constraint, at reaching ! of the time constraint, etc.).

[0100] At operation 2230 the method 2200 includes the computing system performing a second decoding operation based on the constraint status to generate a second portion (e.g., a second token) of the output. For example, referring to FIG. 3, in processing the input 3110 the one or more machine-learned models 3122 may be configured to generate a second portion (e.g., a second token) which forms part of the output which is eventually generated as the generated content 3130.

[0101] The constraint status after the second portion of the output is decoded may be received (e g., via the token counter 3124 and / or timer 3126) by the encoder 3128 similar to operation 2220. In a further decoding operation subsequent to the second decoding operation (e.g., a next decoding operation, at a predetermined interval, etc.), the one or more machine-learned models 3122 may be configured to perform the further decoding operation based on current constraint status information provided by the encoder 3128. The decoding operations may be repeated until the constraint is reached (e.g., n times). For example, the value of nmay correspond to a number of decoding operations until the constraint is reached (e.g., n may equal 500 decoding operations when the constraint is associated with an output corresponding to a maximum number of tokens of 500, or the value of n may equal 1000 decoding operations when the constraint is associated with a maximum latency of 1 second and each decoding operation takes an average of 1 ms, or the value of n may be unknown beforehand and the decoding operations can be repeated until the specified latency is reached based on the constraint status). Upon reaching the constraint, the one or more machine-learned models 3122 may be configured to generate an end of sequence token to prevent the further prediction of tokens and to indicate that the generated output is complete.

[0102] At operation 2240 the method 2200 includes the computing system receiving a constraint status after a n-lth portion of the output is decoded. For example, in some implementations after generating the n-lth portion (e.g., the n-lth token), the content generation application 3120 may be configured to track or monitor a status of the constraint via the token counter 3124 and / or timer 3126 as explained with respect to operation 2220. The constraint status may be provided to the encoder 3128. The encoder 3128 may be configured to provide the constraint status as an input to the one or more machine-learned models 3122 for a next or subsequent decoding operation. For example, the encoder 3128 may be configured to encode the constraint status as a soft token and provide the soft token to the one or more machine-learned models 3122. In some implementations, the constraint status information (e g., the soft token) may be passed to the one or more machine-learned models 3122 as another token that attends to all other tokens or to a subset of the tokens, e g. the most recent ones. In some implementations, the constraint status information (e.g., the soft token) may be passed to the one or more machine-learned models 3122 with a skip connection to feed-forward layers, thus reducing attention cost.

[0103] At operation 2250 the method 2200 includes the computing system performing a nth decoding operation based on the constraint status to generate the nth portion (e g., a nth token) of the output. For example, referring to FIG. 3, in processing the input 3110 the one or more machine-learned models 3122 may be configured to generate a nth portion which forms part of the output which is eventually generated as the generated content 3130, similar to operation 2230. The decoding operation at operation 2250 may correspond to a last decoding operation before generating a final output for responding to the input 3110 (query). In some implementations, the last decoding operation may correspond to a decodingoperation at which the constraint is reached (e.g., a maximum token number and / or a maximum latency for generating the output).

[0104] At operation 2260 the method 2200 includes the computing system generating a final output based on the generated portions for responding to the query. For example, after the last decoding operation is performed (e g., as described wi th respect to operation 2250), a final result can be presented to the user in response to the query', for example, as generated content 3130. For example, the generated portions (which may correspond to tokens) may be decoded back into text using a tokenizer to form the final output. For example, the generated content 3130 may be presented visually, audibly, or combinations thereof.

[0105] As an example implementation, the computing system may receive a query from a user such as ‘‘Explain the rules of soccer to me in 500 words." The one or more machine-learned models 3122 may be configured to analyze (e.g., parse) the input to determine a constraint on the output (e.g., 500 words) included in the query and to identify an intent or goal of the query (e.g.. explain the rules of soccer), for example, by classifying the tokens corresponding to the query and utilizing contextual embeddings. In some implementations, the constraint configuration can be given as a string or in a structured format (e.g., JSON) and the encoder 3128 may be configured to encode a constraint status (e.g., as a soft token). The soft token may have an embedding with the same shape as other embeddings of the vocabulary of the one or more machine-learned models 3122.

[0106] For example, the computing system may be configured to track a status of the constraint (e.g., via the token counter 3124 and encoder 3128) while performing decoding operations to generate tokens which form the final output. At each decoding operation (or at specified or predetermined decoding operations), the one or more machine-learned models 3122 may receive information (e.g., the soft token) from the encoder 3128 indicating the constraint status. As described herein, the constraint status may indicate a number of tokens generated thus far, a time spent generating the output thus far, a number of tokens remaining based on the tokens generated thus far and the maximum number of tokens indicated by the constraint, a time remaining to generate the output based on the time spent generating the output thus far and the maximum time for generating the output indicated by the constraint, etc. In some implementations, other information associated with the constraint status can be provided and encoded as part of the soft token. For example, the constraint status can indicate a time spent for each decoding operation, an average time spent on each decoding operation, an average time that it takes to obtain information (e.g., via a call operation) froman external tool (e.g., via an API) for performing a decoding operation (e.g., 2 seconds), etc. Therefore, the one or more machine-learned models 3122 are given information during the decoding process that makes the one or machine-learned models 3122 aware of how much time is left and / or how many tokens remain to provide a final output that is complete and cohesive. In some implementations, the content generation application 3120 (the one or more machine-learned models 3122) may include a corrector network that is configured to update the encoding of the constraint status for more efficiently updating a stale embedding.

[0107] The one or more machine-learned models 3122 may be configured to repeatedly generate a next portion (e.g., next token) forming the output, based on the current or updated constraint status fed to the one or more machine-learned models 3122 by the encoder 3128, until the constraint is reached or until the one or more machine-learned models 3122 determines a full response to the query has been generated and the constraint has not yet been reached.

[0108] Configuring the one or more machine-learned models 3122 to respect a limit or constrain the number of tokens generated for an output (e.g., based on a user input or user-defined constraint or computing system defined constraint regarding a maximum size of the output and / or a maximum latency) while monitoring a status of the constraint during a decoding process, can conserve or save processing power, energy', and fees associated with generating the output, while also satisfying expectations of a user or computing system by ensuring that a complete response is generated before an end of sequence token is generated. Controlling latency during inference time for the one or more machine-learned models 3122 can also be applied and can further take into account circumstances where the one or more machine-learned models 3122 perform external tool and API calls such that bandwidth and network resources can be efficiently utilized. Accordingly, the one or more machine-learned models 3122 are configured to honor token and / or time constraints described in a prompt by making the one or more machine-learned models 3122 aware of the number of decoding operations remaining and / or of the time remaining for providing an output. The one or more machine-learned models 3122 can include one or more of sequence models, large language models, generative machine-learned models, neural networks, etc. In some implementations, for an image content generation task or a video content generation task, the decoding operations described with respect to FIGS. 2A-2B may not decode token by token but instead may be modified by implementing a diffusion model that outputs a noisy version of the image (or frame) initially and then iteratively performs decoding operations that comprise denoising operations to iteratively refine the output until the constraint is reached. Forexample, the number of iterations or a latency for content generation may be indicated or defined by the constraint provided by the query or associated with the query, and the constraint status can be updated at each decoding operation (and / or at certain timepoints) and fed as an input to the one or more machine-learned models 3122 via an encoder as described herein with respect to encoder 3128 and FIGS. 2A-3.

[0109] Referring to FIG. 4, the computing system (e.g., the content generation application 3120 or a separate training application), may be configured to train the one or more machine-learned models (e.g.. the one or more machine-learned models 3122) to generate content which respects one or more constraints included in the query from a user and / or one or more constraints specified by the computing system, based on the one or more training examples which comprise a query7and one or more constraints.

[0110] The flow diagram of FIG. 4 illustrates an example method for implementing aspects of the disclosure. Although shown in a particular sequence or order, unless otherwise specified, the order of the processes can be modified. Thus, the illustrated embodiments should be understood only as examples, and the illustrated processes can be performed in a different order, and some processes can be performed in parallel. Additionally, one or more processes can be omitted in various embodiments. Thus, not all processes are required in every embodiment. Other process flows are possible.[Oil 1] The operations of FIG. 4 will be explained with reference to FIGS. 3, 5A, and 5B. FIGS. 3, 5A, and 5B illustrate example block diagrams or architectures of environments for training or implementing one or more machine-learned models, according to one or more example embodiments of the disclosure.

[0112] Referring to FIG. 4, at operation 4110 the method 4100 includes a computing system (e.g., the content generation application 132, content generation application 332, content generation application 3120, and / or one or more separate training applications, etc.) obtaining a query and one or more constraints. For example, referring to FIG. 5A, computing system 5100 may include the one or more machine-learned models 5120 (which can correspond to the one or more machine-learned models 134, one or more machine-learned models 334, one or more machine-learned models 3122, etc.) which are configured to receive the query and one or more constraints 5110 to train the one or more machine-learned models 5120 to generate model-predicted generated content 5130 that respects the one or more constraints as described herein with respect to FIGS. 2A-2B and 3.

[0113] Referring to FIG. 4, at operation 4120 the method 4100 includes the computing system processing a query to generate model-predicted generated content. The query’ can include a request to generate an output (e.g., code, a textual output, an image, a video, etc.) while satisfying one or more constraints included in the query or associated with the query. An example query may include a request for generating code to perform a task in a computing program that can be generated by the one or more machine-learned models within a certain time period, a request to generate a summary that is less than or equal to a specified length, etc. For example, referring to the computing system 5100 of FIG. 5 A, the one or more machine-learned models 5120 may be configured to process the query’ and one or more constraints 5110 to generate model-predicted generated content 5130 based on the content of the query and the one or more constraints (e.g., in a manner as described with respect to FIGS. 2A-2B). In some implementations, the computing system may be configured to select or utilize a loss function 5140 to evaluate the performance of the one or more machine-learned models 5120 based on the ground truth training examples 5160. For example, the one or more training examples 5160 can serve as training data (e.g., datasets) that reflect generated content which respects constraints given in a query or in association with a query. In some implementations, the one or more training examples 5160 can be labeled (e.g., according to the type or genre of the query', a subject or topic of the query’, a kind of task associated with the query such as generating code, an image, a video, etc., according to a type or number of constraints, etc.). In some implementations, the computing system may include an optimizer configured to improve the performance of the one or more machine-learned models 5120 to minimize the loss function (e.g., by updating the parameters of the one or more machine-learned models 5120).

[0114] For example, the one or more machine-learned models 5120 can be trained by providing different inputs (different queries) including different constraints and the output of the one or more machine-learned models 5120 can be scored using a reward model. For example, the reward model may be configured to reward or penalize the one or more machine-learned models 5120 based on yvhether the output corresponds to the expected output (e.g., ground truth output) and yvhether the output satisfies the one or more constraints.

[0115] At operation 4130 the method 4100 includes the computing system evaluating the model-predicted generated content and yvhether the one or more constraints are satisfied based on ground truth training examples and computing the loss (e.g., according to the loss function). At operation 4140 the method 4100 includes the computing system training theone or more machine-learned models based on the computed loss. For example, referring to the computing system 5100 of FIG. 5 A, the one or more machine-learned models 5120 may be configured to evaluate the model-predicted generated content 5130 and whether the one or more constraints are satisfied based on ground truth training examples 5160 and to compute the loss (e.g., according to the loss function 5140), which is provided as a loss function output 5150 to the one or more machine-learned models 5120. The loss function 5140 may be implemented to measure the difference (or error) between the one or more machine-learned model's 5120 predicted outputs and the true or expected outputs (ground truth). For example, the true or expected outputs (ground truth) may correspond to the generated content and whether the one or more constraints were satisfied that are associated with a particular training example, which can be obtained from the ground truth training examples 5160. The loss function output 5150 of the loss function 5140 can indicate the performance of the one or more machine-learned models 5120 and can be input to the one or more machine-learned models 5120 for further training of the one or more machine-learned models 5120. For example, the one or more machine-learned models 5120 can be further trained or optimized by adjusting model parameters and / or weights to minimize the loss function output 5150. In some implementations, the computing system may be configured to evaluate the model-predicted generated content (which can be provided as the output generated content 3130), to compute the loss (e.g., according to a loss function) and to train the one or more machine-learned models 3122 based on the computed loss (e.g., by adjusting model parameters and / or weights to minimize the loss function output).

[0116] In some implementations, if the one or more machine-learned models 5120 do not output an end-of-sequence (EOS) token before a constraint was reached, the computing system can penalize the one or more machine-learned models 5120. For example, the penalty may be proportional to the number of additional tokens in excess of the constraint or proportional to the time needed to generate the output in excess of the time constraint, or the penalty may be non-linear (e.g., a quadratic error). If the one or more machine-learned models 5120 output the EOS token before the constraint was reached, the computing system and / or a user may analyze the output sequence to see whether the output is both coherent and completely answers the query. Evaluating the output can be done automatically (e.g., via an autorater) or by having the user provide feedback (e.g., via RLHF). If the output was not coherent or did not fully answer the query, the reward model or user can provide a penalty to the machine-learned model.

[0117] In some implementations, the one or more machine-learned models 5120 may be finetuned according to the operations of FIG. 4 (e.g., training an existing content generation model specifically on a generated dataset comprising training examples that are directed to a particular genre, category, topic, subject, type of constraint, magnitude of constraint, etc.), allowing the one or more machine-learned models 5120 to adapt and specialize in the representation of the specific query and / or constraint captured in the dataset. In some implementations, the one or more machine-learned models 5120 may be trained according to the operations of FIG. 4 by combining a newly generated dataset comprising the training examples as described herein, with other relevant training datasets. For example, the entire content generation model can be trained from scratch using the dataset, allowing the model to learn from a broader range of examples, potentially leading to a more generalized understanding of content generation tasks while still incorporating the unique characteristics of the new dataset.

[0118] For example, referring to the computing system 5200 of FIG. 5B, the one or more machine-learned models 5220 may be configured to receive a query 5210 which can include one or more constraints or be associated with one or more constraints. In some implementations, additional context information can be provided as an input to the one or more machine-learned models 5220. For example, the additional context information can include position information, environmental information, biometric information, academic information, etc., associated with the computing system and / or one or more users associated with the computing system, which can be utilized by the one or more machine-learned models 5220 for outputting the generated content 5230. Referring back to FIG. 3, the one or more machine-learned models 3122 may be configured to receive an input 3110 which includes the query and one or more constraints. In some implementations, the one or more machine-learned models 3122 may be configured to receive the additional context information as part of external content 3140.

[0119] The one or more machine-learned models 5220 may be configured to interpret the query (e.g., through parsing the input via natural language processing) and to identify key objects, actions, environments, etc., from the query, and to determine various conditions or constraints associated with the content to be generated (e.g., content length, time for output generation, etc.). The one or more machine-learned models 5220 may be configured to implement a text-to-text model, a text-to-image model, and / or text-to-video model to generate the generated content 5230. The one or more machine-learned models 5220 may beconfigured to generate other features including audio, visual effects, etc., and to render the final output content (e.g., by post-processing, formatting, etc.). In some implementations, for an image content generation task or a video content generation task, the decoding operations described with respect to FIGS. 2A-2B may not decode token by token but instead may be modified by implementing a diffusion model that outputs a noisy version of the image (or frame) initially and then iteratively performs decoding operations that comprise denoising operations to iteratively refine the output until the constraint is reached. For example, the number of iterations or a latency for content generation may be indicated or defined by the constraint provided by the query' or associated with the query', and the constraint status can be updated at each decoding operation (and / or at certain timepoints) and fed as an input to the one or more machine-learned models 5220 via an encoder as described herein with respect to encoder 3128 and FIGS. 2A-3.

[0120] Referring to FIG. 5B, in some implementations, the computing system 5200 may include a Retrieval-Augmented Generation (RAG) repository 5240 which is configured to store training examples 5250 that can be referenced by the one or more machine-learned models 5220 when the one or more machine-learned models 5220 generates the generated content 5230 (or more generally an output) in response to or based on receiving the query 5210 including the one or more constraints. For example, the one or more machine-learned models 5220 may be configured to retrieve or find relevant information from the RAG repository' 5240 for generating an output by integrating external knowledge from the RAG repository 5240. That is, the one or more machine-learned models 5220 may be configured to retneve the most relevant pieces of information from the training examples stored in the RAG repository 5240 to provide a more informed, contextually7accurate output. The computing system may be configured to store the one or more training examples in a retrieval-augmented generation (RAG) repository 5240. the one or more machine-learned models 5220 may be configured to receive the query 5210 (e.g., to generate content that adheres to one or more constraints regarding the output) and to process the query7by referencing the one or more training examples in the RAG repository' 5240, to generate an output responsive to the query 5210. Machine-learned models other than those trained as described with respect to FIG. 4 can also reference the RAG repository 5240 for training examples that can be quickly searched and retrieved by the machine-learned model when needed to enhance its response generation capabilities. The machine-learned model can then generate more accurate and contextually relevant responses as the RAG repository 5240storing the training examples can serve as a lookup tool for up-to-date information beyond the machine-learned model’s current configuration.

[0121] FIG. 6 depicts a flowchart of a method 6000 for training one or more machine-learned models according to aspects of the disclosure. For instance, an example machine-learned model can include one or more of a LLM, a generative machine-learned model, etc. For example, the one or more machine-learned models may be configured to generate some or all of the content based on receiving a query', as described herein.

[0122] FIG. 6 is a flow chart diagram illustrating another example method for training a machine-learned model according to example implementations of aspects of the disclosure. One or more portion(s) of example method 6000 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 drawings. Each respective portion of example method 6000 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 6000 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. FIG. 6 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 disclosure. FIG. 6 is described with reference to elements / terms described with respect to other systems and drawings for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 6000 can be performed additionally, or alternatively, by other systems.

[0123] At 6002, example method 6000 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 6000 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 disclosure.

[0124] At 6004, example method 6000 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can be directly obtained from the one or more machine-learned models or can be a downstream result of a chain of processing operations that includes an output of the one or more machine-learned models.

[0125] At 6006, example method 6000 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).

[0126] At 6008, example method 6000 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 6000 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0127] In some implementations, example method 6000 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 profile, such as based on accuracy, precision, recall, etc.).

[0128] In some implementations, example method 6000 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 6000 can be implemented for pre-training a machine-learned model. Pre-training can include, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety' of tasks / data types. In some implementations, example method 6000 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)). An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use.Example Machine-Learned Models

[0129] FIG. 7 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.

[0130] 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 networks, linear regression models, k-means clustering models, etc.

[0131] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recunent 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.

[0132] 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 an ensemble of different models that can cooperatively interact to process data from input(s) 2. For example, 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, ARXIV:2202.09368V2 (Oct. 14, 2022).

[0133] Input(s) 2 can generally include or otherwise represent various types of data. Input(s) 2 can include one type or many different types of data. Output(s) 3 can be data of the same type(s) or of different types of data as compared to input(s) 2. Output(s) 3 can include one type or many different t pes of data.

[0134] Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g., binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer's central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like. Data can be raw or processed and can be in any format or schema.

[0135] 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.

[0136] An example input 2 can include one or multiple data types, such as the example data types noted above. An example output 3 can include one or multiple data types, such as the example data types noted above. The data type(s) of input 2 can be the same as 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 disclosure are not limited to those examples noted above.Example Machine-Learned Sequence Processing Models

[0137] FIG. 8 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.

[0138] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models in the text domain are referred to as “Large Language Models,” or LLMs. See, e.g, PaLM 2 Technical Report, GOOGLE, https: / / ai.google / static / documents / palm2techreport.pdf (n d.). Other example sequence processing models can operate in other domains, such as image domains, see, e.g., Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXIV:2010.11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et al., MusicLM: Generating Music From Text, ARXIV:2301.11325V1 (Jan. 26, 2023), biochemical domains, see. e.g., Jumper et al., Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example. Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), relatively small models (e.g., fewer parameters, computationally lightweight, etc ), or both.

[0139] 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”).

[0140] 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 dow n into pieces that collectively represent the content of the portion of the input data. The pieces can provide the elements of the sequence.

[0141] 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.

[0142] 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 CONFE ENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66-71 (October 31-November 4, 2018). https: / / aclanthology.org / D18-2012.pdf. Imagebased input source(s) can be tokenized by extracting and serializing patches from an image.

[0143] 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 FIG. 11 can be the tokens or can be the embedded representations thereof.

[0144] Prediction layer(s) 6 can predict one or more output elements 7-1, 7-2, . . . , 1-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.

[0145] 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 predictionlayer(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.”

[0146] A transformer is an example architecture that can be used in prediction layer(s) 4. See, e.g., Vaswani et al., Attention Is All YouNeed, ARXIV: 1706.03762v7 (Aug. 2, 2023). A transformer is an example of a machine-learned model architecture that uses an attention mechanism to compute associations between items within a context window. The context window can include a sequence that contains input sequence 5 and potentially one or more output element(s) 7-1, 7-2, . . . , 7-N. A transformer block can include one or more attention layer(s) and one or more post-attention layer(s) (e g., feedforward layer(s), such as a multilayer perceptron).

[0147] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as w ell as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information.

[0148] 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.

[0149] 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.

[0150] 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.

[0151] 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).

[0152] 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.

[0153] FIG. 9 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 modality10-3 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.[0154j 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.

[0155] For example, elements 8-0, . . . , 8-9 can indicate particular locations within a multidimensional embedding space. Some elements can map to a set of discrete locations in the embedding space. For instance, elements that correspond to discrete members of a predetermined vocabulary of tokens can map to discrete locations in the embedding space that are associated with those tokens. Other elements can be continuously distributed across the embedding space. For instance, some data types can be broken down into continuously defined portions (e.g., image patches) that can be described using continuously distributed locations within the embedding space.

[0156] 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, ahigh-order embedding space can encode information that can be independent of data modalities in which the information is expressed.

[0157] 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.

[0158] 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).

[0159] 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.).

[0160] 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.Example Machine-Learned Model Development Platform

[0161] FIG. 10 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 developmentplatform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0162] 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.

[0013] 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.

[0164] 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.

[0165] 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).

[0166] 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.

[0167] 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., de-noising, 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.

[0168] 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 fine-tune development model 16.

[0169] Prompt libraries 17-4 can include sets of inputs configured to induce behavior aligned with desired performance criteria. Prompt libraries 17-4 can include few-shot prompts (e.g., inputs providing examples of desired model outputs for prepending to a desired runtime query), chain-of-thought prompts (e.g., inputs providing step-by-step reasoning within the exemplars to facilitate thorough reasoning by the model), and the like.

[0170] 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.

[0171] 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).

[0172] 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.

[0173] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine-learned models. In this manner, for instance, a first model can process information about a task and output an input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0174] 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.

[0175] Although various training examples described herein with respect to model development platform 12 refer to "pre-training" and "fine-tuning." it is to be understood that model alignment toolkit 17 can generally support a wide variety of training techniques adapted for training a wide variety of machine-learned models. Example training techniques can correspond to the example training method 6000 described above.

[0176] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance qualify 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 (based on) 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.

[0177] 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’').

[0178] 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.

[0179] 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.

[0180] 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.

[0181] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardw are 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, low er unified memory requirements, etc. Tools for distillation 19-3 can provide for the training of lighter- eight 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 inresource-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.

[0182] 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.

[0183] FIG. 11 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other drawings. 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. 11 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 disclosure. FIG. 11 is described with reference to elements / terms described with respect to other systems and drawings for exemplary illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0184] 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.

[0185] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pre-training stage 22 can be implemented using one or more pre-training pipelines 17-2 over data fromdataset(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).

[0186] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as anew 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 satis factory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0187] Fine-tuned model 29 can then be anew 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.

[0188] In some implementations, computational optimization operations can be applied before, during, or after each stage. For instance, initialized model 21 can undergo computational optimization 29-1 (e.g., using computational optimization toolkit 19) before pre-training stage 22. Pre-trained model 23 can undergo computational optimization 29-2 (e.g., using computational optimization toolkit 19) before fine-tuning stage 24. Fine-tuned model 25 can undergo computational optimization 29-3 (e.g., using computational optimization toolkit 19) before refinement with user feedback 26. Refined model 27 can undergo computational optimization 29-4 (e.g., using computational optimization toolkit 19) before output to downstream system(s) 28. Computational optimization(s) 29-1, . . . , 29-4 can all be the same, all be different, or include at least some different optimization techniques.Example Machine-Learned Model Inference System

[0189] FIG. 12 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.

[0190] 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.

[0191] 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.

[0192] 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.

[0193] For example, model host 31 can operate on a server system that provides a machinelearning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-areanetwork). 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.

[0194] 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.

[0195] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include weights or other model components that are stored in persistent storage, temporarily cached, or loaded into high-speed memory. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include 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.

[0196] Compute resource(s) 31-2 can include one or more processors (central processing units, graphical processing units, tensor processing units, machine-learning accelerators, etc.) connected to one or more memory devices. Compute resource(s) 31-2 can include a dynamic pool of available resources shared with other processes. Compute resource(s) 31-2 can include memory devices large enough to fit an entire model instance in a single memory instance. Compute resource(s) 31-2 can also shard model instance(s) across multiple memory devices (e.g., using data parallelization or tensor parallelization, etc.). This can be done to increase parallelization or to execute a large model using multiple memory devices which individually might not be able to fit the entire model into memory.

[0197] 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.

[0198] 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.

[0199] 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 arnve 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.

[0200] 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.

[0201] 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 otherw ise 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 imagedata 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.

[0202] 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 image processing output defines, for each pixel of one of the input images, a motion of the scene depicted at the pixel between the images in the network input.

[0203] 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 naturallanguage 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).

[0204] 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.

[0205] 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 anotherexample, machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.

[0206] 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.

[0207] 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) 1 can process the sensor data to generate a visualization output. As another example, machine-learned model(s) 1 can process the sensor data to generate a diagnostic output. As another example, machine-learned model(s) 1 can process the sensor data to generate a detection output.

[0208] In some implementations, machine-learned model(s) 1 can be configured to perform a task that includes encoding input data for reliable and / or efficient transmission or storage (and / or corresponding decoding). For example, the task may be an audio compression task. The input may include audio data and the output may comprise compressed audio data. In another example, the input includes visual data (e.g., one or more images or videos), the output comprises compressed visual data, and the task is a visual data compression task. In another example, the task may comprise generating an embedding for input data (e.g., input audio or visual data). In some cases, the input includes audio data representing a spoken utterance and the task is a speech recognition task. The output may comprise a text outputwhich 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.

[0209] 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.

[0210] 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.

[0211] 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 process and accomplish steps toward accomplishing the requested functionality. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of performing a function. Multiple steps can be performed, with a final output being obtained that is responsive to the initial instructions.

[0212] 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.

[0213] 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) 1 can be configured to generate pixel data of an image. Values for channel(s) associated with the pixels in the pixel data can be selected based on the context (e.g., based on a probability determined based on the context).

[0214] 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).

[0215] 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).Example Computing Systems and Devices

[0216] FIG. 13 is a block diagram of an example networked computing system that can perform aspects of example implementations of the 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 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 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 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 disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.).

[0217] Network 49 can be any type of communications network, such as a local area network (e.g., intranet), wide area netw ork (e.g., Internet), or some combination thereof and caninclude 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 FIG. 13 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0218] 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).

[0219] 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.

[0220] 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.

[0221] 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.

[0222] 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.

[0223] 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.

[0224] Server computing system 60 can store or otherwise include one or more machine-learned models 65. Machine-learned model (s) 65 can be the same as or different from machine-learned model(s) 55. Machine-learned models 65 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 65 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 65 can be received from computing device 50, model development platform system 70, third party system(s) 80, or developed locally on server computing system(s) 60. Machine-learned model(s) 65 can be loaded into memory 62 and used or otherwise implemented byprocessor(s) 61. Server computing system(s) 60 can implement multiple parallel instances of machine-learned model(s) 65.

[0225] In an example configuration, machine-learned models 65 can be included in or otherwise stored and implemented by server computing system 60 to establish a client-server relationship with computing device 50 for serving model inferences. For instance, server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50. For instance, machine-learned models 65 can be implemented by server computing system 60 as a portion of a web service (e.g., remote machine-learned model hosting service, such as an online interface for performing machine-learned model operations over a network on server computing system(s) 60). For instance, server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection. For instance, computing device 50 can be a workstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50. Machine-learned models 65 can work cooperatively or interoperatively with machine-learned models 55 on computing device 50 to perform various tasks.

[0226] 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.

[0227] 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 includeone or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 82 can store data 83 and instructions 84 which can be executed by processor(s) 81 to cause third-party system(s) 80 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein wi th 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).

[0228] FIG. 13 illustrates one example arrangement of computing systems that can be used to implement the 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).

[0229] FIG. 14 is a block diagram of an example computing device 98 that performs according to example embodiments of the 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 navigation application, a content generation application, a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, a social media application, a chat application, etc. As illustrated in FIG.14, 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 cancommunicate 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.

[0230] FIG. 15 is a block diagram of an example computing device 99 that performs according to example embodiments of the 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 navigation application, a content generation application, a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, a social media application, a chat 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).

[0231] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in FIG. 15, 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.

[0232] 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 FIG. 15, 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).Additional Disclosure

[0233] The technology7discussed 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 agreat 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.

[0234] 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 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 disclosure 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.”

[0235] Terms used herein are used to describe the example embodiments and are not intended to limit and / or restrict the disclosure. The singular forms “a,” “an” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. In this disclosure, terms such as "including", "having", “comprising”, and the like are used to specify features, numbers, steps, operations, elements, components, or combinations thereof, but do not preclude the presence or addition of one or more of the features, numbers, steps, operations, elements, components, or combinations thereof.

[0236] The term "and / or" includes a combination of a plurality- of related listed items or any item of the plurality of related listed items. For example, the scope of the expression or phrase "A and / or B" includes the item "A", the item "B", and the combination of items "A and B”.

[0237] In addition, the scope of the expression or phrase "at least one of A or B" is intended to include all of the following: (1) at least one of A, (2) at least one of B, and (3) at least one of A and at least one of B. Likewise, the scope of the expression or phrase "at least one of A, B, or C" is intended to include all of the following: (1) at least one of A, (2) at least one of B, (3) at least one of C, (4) at least one of A and at least one of B, (5) at least one of A and at least one of C, (6) at least one of B and at least one of C, and (7) at least one of A, at least one of B, and at least one of C.

[0238] It will be understood that, although the terms first, second, third, etc., may be used herein to describe various elements, the elements are not limited by these terms. Instead, these terms are used to distinguish one element from another element. For example, without departing from the scope of the disclosure, a first element may be termed as a second element, and a second element may be termed as a first element.

[0239] 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 disclosure.

[0240] 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 vanous implementations, X might be unable to perform Y and remain within the scope of the disclosure.

[0241] To the extent terms including "module", and "unit," and the like are used herein, these terms may refer to, but are not limited to, a software or hardware component or device, such as a Field Programmable Gate Array (FPGA) or Application Specific Integrated Circuit (ASIC), which performs certain tasks. A module or unit may be configured to reside on an addressable storage medium and configured to execute on one or more processors. Thus, a module or unit may include, by way of example, components, such as software components,object-oriented software components, class components and task components, processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. The functionality provided for in the components and modules / units may be combined into fewer components and modules / units or further separated into additional components and modules.

[0242] Aspects of the above-described example embodiments may be recorded in non-transitory computer-readable media including program instructions to implement various operations embodied by a computer. The media may also include, alone or in combination with the program instructions, data files, data structures, and the like. Examples of non-transitory computer-readable media include magnetic media such as hard disks, floppy disks, and magnetic tape; optical media such as CD ROM disks, Blu-Ray disks, and DVDs; magneto-optical media such as optical discs; and other hardware devices that are specialty configured to store and perform program instructions, such as semiconductor memory, readonly memory (ROM), random access memory (RAM), flash memory, USB memory, and the like. Examples of program instructions include both machine code, such as produced by a compiler, and files containing higher level code that may be executed by the computer using an interpreter. The program instructions may be executed by one or more processors. The described hardware devices may be configured to act as one or more software modules in order to perform the operations of the above-described embodiments, or vice versa. In addition, a non-transitory computer-readable storage medium may be distributed among computer systems connected through a network and computer-readable codes or program instructions may be stored and executed in a decentralized manner. In addition, the non-transitory computer-readable storage media may also be embodied in at least one application specific integrated circuit (ASIC) or Field Programmable Gate Array (FPGA).

[0243] Each block of the flowchart illustrations may represent a unit, module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks may occur out of order. For example, two blocks shown in succession may7in fact be executed substantially7concurrently (simultaneously) or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

[0244] Further to the descriptions above, a user may be provided with controls allowing the user to make an election as to both if and when systems, programs, or features describedherein may enable collection of user information (e.g., information about a user’s social network, social actions, or activities, profession, a user’s preferences, or a user’s current location), and if the user is sent content or communications from a server. In addition, certain data may be treated in one or more ways before it is stored or used, so that personally identifiable information is removed. For example, a user's identity may be treated so that no personally identifiable information can be determined for the user, or a user’s geographic location may be generalized where location information is obtained (such as to a city, ZIP code, or state level), so that a particular location of a user cannot be determined. Thus, the user may have control over what information is collected about the user, how that information is used, and what information is provided to the user.

[0245] While the disclosure has been described with respect to various example embodiments, 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 disclosure does not preclude inclusion of such modifications, variations and / or additions to the disclosed subject matter as would be readily apparent to one of ordinary skill in the art. For example, 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 disclosure covers such alterations, variations, and equivalents.

Claims

WHAT IS CLAIMED IS:

1. A computer-implemented method, comprising:receiving, by a computing system comprising one or more machine-learned models, a query which includes a constraint; andprocessing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes:performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output,receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, andperforming, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

2. The computer-implemented method of claim 1, further comprising: encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

3. The computer-implemented method of claim 2, whereinthe first portion includes one or more first soft tokens, andthe second portion includes one or more second soft tokens.

4. The computer-implemented method of claim 3, whereinthe one or more machine-learned models perform a plurality of decoding operations to generate a plurality of tokens which form at least part of the output, andthe one or more machine-learned models receive an update to the constraint status after each token is generated according to a respective decoding operation among the plurality of decoding operations.

5. The computer-implemented method of claim 3, whereinthe one or more machine-learned models perform a plurality of decoding operations to generate a plurality’ of tokens which form at least part of the output, andthe one or more machine-learned models receive an update to the constraint status after a predetermined number of tokens are generated according to a predetermined number of decoding operations among the plurality of decoding operations.

6. The computer-implemented method of claim 3, whereinthe one or more soft tokens are provided to the one or more machine-learned models via a skip connection to one or more feed-forward layers of the one or more machine-learned models.

7. The computer-implemented method of claim 1, whereinthe constraint indicates a maximum number of tokens forming the output, and the one or more machine-learned models receive an update to the constraint status after one or more predetermined proportions of the maximum number of tokens are generated according to one or more decoding operations performed by the one or more machine-learned models among a plurality’ of decoding operations performed by the one or more machine-learned models.

8. The computer-implemented method of claim 1, whereinthe constraint indicates a maximum time to generate the output, andthe constraint status indicates a remaining time to generate the output.

9. The computer-implemented method of claim 8, whereinthe one or more machine-learned models perform a plurality- of decoding operations to generate a plurality of tokens which form at least part of the output, andthe one or more machine-learned models receive an update to the constraint status indicating the remaining time to generate the output after each token is generated according to a respective decoding operation among the plurality’ of decoding operations.

10. The computer-implemented method of claim 8, whereinthe one or more machine-learned models receive an update to the constraint status after one or more predetermined proportions of the maximum time to generate the output according to one or more decoding operations performed by the one or more machine-learned models among a plurality of decoding operations performed by the one or more machine-learned models.

11. The computer-implemented method of claim 8, wherein the one or more machine-learned models receive an update to the constraint status after one or more predetermined intervals of time elapse during decoding operations performed by the one or more machine-learned models among a plurality of decoding operations performed by the one or more machine-learned models.

12. The computer-implemented method of claim 1, whereinthe constraint indicates a maximum time to generate the output, andthe constraint status indicates a remaining time to generate the output and an elapsed time that the one or more machine-learned models has spent performing decoding operations in generating the output, andthe method further comprises:encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

13. The computer-implemented method of claim 1, whereinthe constraint indicates a maximum number of tokens forming the output, and the constraint status indicates a remaining number of tokens before the maximum number of tokens is reached and a number of tokens the one or more machine-learned models has currently output by performing decoding operations in generating the output, and the method further comprises:encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

14. A computing system, comprising:one or more processors; andone or more non-transitory computer-readable media storing:one or more machine-learned models; andinstructions that, when implemented, cause the one or more processors to perform operations, the operations comprising:receiving a query which includes a constraint; and processing, by the one or more machine-learned models, the query to generate an output which satisfies the constraint, wherein the processing includes:performing, by the one or more machine-learned models, a first decoding operation to generate a first portion of the output,receiving, by the one or more machine-learned models, a constraint status after the first portion of the output is decoded, andperforming, by the one or more machine-learned models, a second decoding operation based on the constraint status, to generate a second portion of the output.

15. The computing system of claim 14, wherein the operations further comprise: encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

16. The computing system of claim 15, whereinthe first portion includes one or more first soft tokens, andthe second portion includes one or more second soft tokens.

17. The computing system of claim 16, whereinthe one or more machine-learned models perform a plurali ty of decoding operations to generate a plurality of tokens which form at least part of the output, andthe one or more machine-learned models receive an update to the constraint status after each token is generated according to a respective decoding operation among the plurality of decoding operations.

18. The computing system of claim 14, whereinthe constraint indicates a maximum time to generate the output, andthe constraint status indicates a remaining time to generate the output and an elapsed time that the one or more machine-learned models has spent performing decoding operations in generating the output, andthe operations further comprise:encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

19. The computing system of claim 14, whereinthe constraint indicates a maximum number of tokens forming the output, andthe constraint status indicates a remaining number of tokens before the maximum number of tokens is reached and a number of tokens the one or more machine-learned models has currently output by performing decoding operations in generating the output, and the operations further comprise:encoding the constraint status as one or more soft tokens; andproviding the one or more soft tokens to the one or more machine-learned models.

20. A computer-implemented method, comprising:obtaining, by a computing system comprising one or more processors and one or more machine-learned models, a query' and one or more constraints associated with the query'; processing, by the one or more machine-learned models, the query to generate a model-predicted generated content;evaluating the model-predicted generated content based on one or more ground truth training examples and based on whether the model-predicted generated content satisfied the one or more constraints, to determine a loss function; andtraining, by the computing system, the one or more machine-learned models based on the loss function.

Images