Semantic clustering for unlimited context window sizes for sequence processing models

Abstract

Systems and methods are provided for semantic clustering for arbitrarily long context windows for machine-learned sequence processing models. A computing system can obtain a context sequence. The computing system can determine a plurality of subsequences of the context sequence. The computing system can determine, using a machine-learned semantic embedding model, a semantic embedding for each subsequence. The computing system can determine, based on the semantic embedding, a plurality of semantic clusters. The computing system can generate, using a machine-learned sequence generation model and based at least in part on the semantic clusters, an output sequence.

Description

SEMANTIC CLUSTERING FOR UNLIMITED CONTEXT WINDOW SIZES FORSEQUENCE PROCESSING MODELSFIELD

[0001] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to semantic clustering for machine-learned sequence processing models.BACKGROUND

[0002] 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 “learn” to generate the desired outputs. The resulting model is often referred to as a machine-learned model.

[0003] In some instances, machine-learned models can be associated with a finite maximum amount of input data the model is configured to process. For example, machine- learned sequence processing models can have a finite context window, wherein inputs longer than a maximum length are effectively truncated, such that part of the input is effectively ignored. However, such a limited input size can in some instances be associated with poorer performance (e.g., inference accuracy) relative to a machine-learned model configured to process more input data. However, increasing a maximum input length can in some instances be associated with a very high computational cost. For example, a computational cost of training a machine-learned sequence processing model can in some instances be proportional to a square of a maximum input length (e.g. O, wherein n is a maximum context length and d is a dimension size of the machine-learned model). For this reason, training a machine- learned model to productively use a very large context window according to prior methods can in some instances be prohibitively expensive.SUMMARY

[0004] Aspects and advantages of embodiments of the present 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 present disclosure provide an example method. In some implementations, the example method can include obtaining, by one or more computing devices, a context sequence. The example method can include determining, by the one or more computing devices, a plurality of subsequences of the context sequence. The example method can include determining, by the one or more computing devices using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence. The example method can include determining, by the one or more computing devices, a plurality of semantic clusters. In the example method, each semantic cluster of the plurality of semantic clusters can include one or more semantic embeddings of the plurality of semantic embeddings. The example method can include generating, by the one or more computing devices using at least one of the first machine-learned sequence processing model and a second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

[0006] The example method can include selecting, by the one or more computing devices based at least in part on the plurality of semantic clusters, one or more subsequences of the plurality of subsequences of the context sequence. The example method can include adding, by the one or more computing devices, data associated with the one or more subsequences to a context window of the second machine-learned sequence processing model. In the example method, the output sequence can be generated based at least in part on the context window.

[0007] In the example method, the data associated with the one or more subsequences can include at least one of the one or more subsequences.

[0008] In the example method, the data associated with the one or more subsequences can include metadata associated with at least one of the one or more subsequences.

[0009] In the example method, the data associated with the one or more subsequences can include a machine-generated summary of at least one of the one or more subsequences.

[0010] In the example method, the data associated with the one or more subsequences can include sequence data generated by the one or more computing devices based at least in part on a semantic embedding associated with at least one of the one or more subsequences.

[0011] In the example method, the data associated with the one or more subsequences can include one or more respective semantic embeddings of the plurality of respective semantic embeddings.

[0012] In the example method, the data associated with one or more subsequences can include sequence data generated by the one or more computing devices based at least in part on a geometric transformation of a semantic embedding associated with at least one of the one or more subsequences.

[0013] In the example method, determining the one or more subsequences can include selecting based on one or more priorities.

[0014] In the example method, selecting based on the one or more priorities can include prioritizing clusters associated with a larger number of semantic embeddings.

[0015] In the example method, selecting based on the one or more priorities can include prioritizing semantic embeddings that are near a center of a semantic cluster.

[0016] In the example method, the center of the semantic cluster can be determined by averaging a plurality of semantic embeddings of the semantic cluster.

[0017] In the example method, the one or more priorities can include a priority determined based on a user input.

[0018] In the example method, selecting based on the one or more priorities can include prioritizing semantic embeddings that are near a semantic embedding associated with a user input.

[0019] In the example method, selecting based on the one or more priorities can include prioritizing semantic embeddings from a cluster having a centroid near a semantic embedding associated with a user input.

[0020] In the example method, selecting based on the one or more priorities can include prioritizing a selection of at least one subsequence from each semantic cluster of the plurality of semantic clusters over the selection of a second subsequence from any semantic cluster of the plurality of semantic clusters.

[0021] In the example method, the first machine-learned model can be characterized by a first number of parameters; the second machine-learned model can be characterized by asecond number of parameters; and the second number can be at least five times larger than the first number.

[0022] In the example method, the second number can be at least 10 times larger than the first number.

[0023] In the example method, the second number can be at least 30 times larger than the first number.

[0024] In the example method, the first machine-learned model can be configured to receive, as input, a sequence comprising multiple tokens; and generate a single semantic embedding based on at least two different tokens of the multiple tokens.

[0025] In the example method, the first machine-learned model can be configured to generate one semantic embedding for each input token it receives. In the example method, determining a semantic embedding associated with a subsequence of the context sequence can include combining a plurality of semantic embeddings generated by the first machine- learned model based on the subsequence.

[0026] In the example method, determining the plurality of semantic clusters can include hierarchical clustering.

[0027] In the example method, at least one semantic cluster of the plurality of semantic clusters can include a plurality of subclusters. In the example method, selecting based on the one or more priorities can include prioritizing a selection of at least one subsequence from each of two or more subclusters of the plurality of subclusters over a selection of a second subsequence from any subcluster of the plurality of subclusters.

[0028] In the example method, at least one semantic cluster of the plurality of semantic clusters can include a plurality of subclusters. In the example method, selecting based on the one or more priorities can include prioritizing semantic embeddings near a centroid of a subcluster of the plurality of subclusters.

[0029] In the example method, determining the plurality of subsequences of the context sequence can include splitting the context based at least in part on a subsequence length threshold.

[0030] In the example method, determining the plurality of subsequences of the context sequence can include splitting the context based at least in part on one or more delimiters or sequence boundaries.

[0031] In the example method, at least one of the one or more delimiters can include one or more numerical, alphabetical, textual, or white-space characters.

[0032] In the example method, determining the plurality of subsequences of the context sequence can include using a machine-learned model.

[0033] The example method can include adding a prompt to the context window of the second machine-learned model. In the example method, the prompt can include one or more instructions or explanations.

[0034] Example aspects of the present disclosure provide one or more example non- transitory computer-readable media storing 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 obtaining a context sequence. The example operations can include determining a plurality of subsequences of the context sequence. The example operations can include determining, using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence. The example operations can include determining a plurality of semantic clusters. In the example operations, each semantic cluster of the plurality of semantic clusters can include one or more semantic embeddings of the plurality of semantic embeddings. The example operations can include generating, using at least one of the first machine-learned sequence processing model and a second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

[0035] Example aspects of the present disclosure provide an example computing system that includes one or more processors and one or more example non-transitory computer-readable media storing 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 obtaining a context sequence. The example operations can include determining a plurality of subsequences of the context sequence. The example operations can include determining, using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence. The example operations can include determining a plurality of semantic clusters. In the example operations, each semantic cluster of the plurality of semantic clusters can include one or more semantic embeddings of the plurality of semantic embeddings. The example operations can include generating, using at least one of the first machine-learned sequence processing model and a second machine- learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

[0036] Other example aspects of the present 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 present disclosure and, together with the description, help explain the related principles.BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 shows a block diagram of an example system according to example implementations of aspects of the present disclosure.

[0038] Figures 2A-2C show illustrations of an example system according to example implementations of aspects of the present disclosure.

[0039] Figure 3 shows a flowchart diagram of an example method according to example implementations of aspects of the present disclosure.

[0040] Figure 4 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0041] Figure 5 is a block diagram of an example processing flow for using machine- learned model(s) to process input(s) to generate output(s) according to example implementations of aspects of the present disclosure;

[0042] Figure 6 is a block diagram of an example sequence processing model according to example implementations of aspects of the present disclosure;

[0043] Figure 7 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example implementations of aspects of the present disclosure;

[0044] Figure 8 is a block diagram of an example model development platform according to example implementations of aspects of the present disclosure;

[0045] Figure 9 is a block diagram of an example training workflow for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0046] Figure 10 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example implementations of aspects of the present disclosure;

[0047] Figure 11 is a block diagram of an example networked computing system according to example implementations of aspects of the present disclosure;

[0048] Figure 12 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure; and

[0049] Figure 13 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure.DETAILED DESCRIPTION

[0050] Generally, the present disclosure is directed to systems and methods for using clusters of machine-learned semantic embeddings with machine-learned sequence processing models. In some instances, provided systems and methods can facilitate the use of unlimited (i.e., arbitrarily long) context sizes by machine-learned models having a finite context window.

[0051] For example, an arbitrarily long input context can be split into semantic chunks (e.g., paragraphs, sentences, lines, etc.). Each chunk can be input to a first sequence processing model (e.g., a lightweight embedding model associated with a low computational cost) to generate one or more machine-learned semantic embeddings associated with the chunk. The semantic embeddings can be clustered using a clustering algorithm to generate a plurality of semantic clusters. In some instances, a first or second machine-learned sequence processing model can synthesize insights (e.g., summaries; identified patterns, topics, or entities; main ideas; suggested questions; etc.) from one or more semantic clusters. Based on the plurality of semantic clusters, one or more semantic chunks and / or synthesized insights can be selected for inclusion in a finite context window of a first, second, or third machine- learned sequence processing model (e.g., a sequence generation model characterized by a higher computational cost than the first sequence processing model). Based on the context selected for inclusion, the sequence processing model can generate one or more outputs (e.g., output sequences).

[0052] Splitting a context into chunks can include a variety of methods. In some instances, a context can be split by one or more document boundaries (e.g., chapters, pages, paragraphs, etc.) or other sequence boundaries (e.g., songs; conversational turns; acts or scenes in an audio or video sequence; etc.). In some instances, a context can be split by one or more delimiters (e.g., white-space characters such as a new-line character; text strings, such as a tag associated with a structured context such as a many-shot prompt; audiodelimiter such as a period of silence; etc.). In some instances, a context can be split into chunks based on one or more chunk length thresholds (e.g., maximum chunk length, minimum chunk length, etc.). In some instances, a context can be split into chunks using a machine-learned model (e.g., based on a machine-learned identification of sequence boundaries, machine-learned identification of sequence similarities, etc.).

[0053] Semantic embedding of a chunk can include using a lightweight sequence processing model configured to generate a semantic embedding based on a token or sequence. In some instances, the model can be configured to generate a semantic embedding for any sequence shorter than a maximum length. In such instances, generating a semantic embedding for a chunk shorter than the maximum length can consist essentially of providing the chunk to the model as input. Generating a semantic embedding for a chunk longer than the maximum length can comprise, for example, splitting the chunk into subsequences that are shorter than the maximum length; generating a semantic embedding for each subsequence; and combining the semantic embeddings (e.g., via average pooling; machine- learned combination; etc.).

[0054] Clustering the semantic embeddings can include any appropriate clustering algorithm (e.g., hierarchical divisive clustering, centroid-based clustering such as k-means, distribution-based clustering such as g-means, connectivity-based clustering such as hierarchical agglomerative clustering, density-based clustering, biclustering, etc.). In some instances, a target number of clusters can be determined based on a size of a finite context window of a sequence generation model used to generate a final output sequence (e.g., in relation to an average length of the context chunks). In some instances, a clustering algorithm (e.g., k-means) can be configured to generate a number of clusters equal to or approximately equal to the target number. In some instances, hierarchical clustering can be used to achieve the target number of clusters. For example, if a first clustering algorithm generates a first number of clusters that is larger than the target number of clusters, then hierarchical agglomerative clustering can be used to merge a plurality of similar clusters into one supercluster. If a first number of clusters is too small, hierarchical divisive clustering can be used. A hierarchical clustering process can be repeated until a target number of clusters is achieved.

[0055] Selecting semantic chunks for inclusion in a finite context window can include, for example, deterministic or probabilistic selection based on one or more priorities. For example, a selection mechanism can be configured to prioritize semantic chunks associated with a large semantic cluster; semantic chunks having a semantic embedding neara center of a semantic cluster or hierarchical semantic subcluster; semantic chunks associated with clusters that are currently unrepresented in the finite context window; semantic chunks having semantic embeddings near a semantic embedding determined based on a user input; etc. In some instances, the selection mechanism can comprise a hierarchical selection mechanism such as a decision tree. In some instances, the selection mechanism can comprise a numerical selection mechanism such as a weighted combination of one or more priorities. For example, in some instances, a selection mechanism can comprise assigning a respective priority score to each chunk of a plurality of semantic chunks, wherein a priority score can comprise, for example, a weighted average of a plurality of priority subscores (e.g., distance from nearest cluster centroid, number of chunks in cluster, etc.).

[0056] As one example, an example selection algorithm can start with a largest semantic cluster of the plurality of semantic clusters and select a subsequence having a semantic embedding closest to a center of the largest semantic cluster. Next, an example selection algorithm can select from the second largest cluster, then the third largest cluster, and so on until either: (a) a context window size limit is met, or (b) at least one respective subsequence has been chosen from each respective semantic cluster of the plurality of semantic clusters. If the context window has room remaining after each semantic cluster is represented in the context window, then a second cluster can be selected from the largest cluster. In some instances, the second cluster can have an embedding that is second closest to a center of the largest semantic cluster, or closest to a center of a largest or second largest hierarchical subcluster of the largest semantic cluster. After each respective semantic cluster is represented in a context window, a respective number of additional subsequences added from a respective cluster or subcluster (e.g., largest cluster, second largest cluster, etc.) can be based on (e.g., approximately proportional to) a size of the respective cluster.

[0057] In another example, an example selection algorithm can start with a semantic cluster having a center that is closest to a semantic embedding of a user input. The example algorithm can select a subsequence from the cluster that is either, for example: (a) closest to a center of the cluster, or (b) closest to the semantic embedding of the user input. Next, the example algorithm can select from the cluster that is second closest to the semantic embedding of the user input, and so on until all clusters are represented (or all clusters within a distance threshold of the semantic embedding of the user input, etc.). After each respective semantic cluster or subcluster is represented in a context window, a respective number of additional subsequences added from a respective cluster or subcluster (e.g., closest cluster toa user input, second closest cluster, etc.) can be based on (e.g., inversely related to) a distance between a center of the respective cluster and the semantic embedding of the user input.

[0058] In some instances, selected semantic chunks can be added directly to a finite context window of a sequence generation model. In some instances, other data associated with the selected semantic chunks can be obtained, and the other data can be added to the finite context window. For example, in some instances, a machine-learned sequence generation model may generate a sequence (e.g., summary, extracted entity or keyword, extracted answer to a question, etc.) based on one or more selected chunks or clusters, and the sequence can be added to the finite context window. In some instances, a computing system can obtain (e.g., receive, retrieve, generate) one or more metadata items associated with the selected chunks (e.g., chapter number, word count, topic, etc.) to add to the finite context window. In some instances, additional data not associated with the selected chunks can be added to the finite context window, such as a prompt (e.g., “Please summarize the following information:”, etc.).

[0059] Systems and methods of the present disclosure provide various technical effects and benefits. In one example aspect, provided systems and methods enable the efficient use of arbitrarily large context sequences for sequence processing models.

[0060] Prior sequence processing models (e.g., transformers) have in some instances been associated with finite context windows. In some instances, a maximum length of a context window can be fixed by a model architecture or training process used to train a sequence processing model. For example, in some instances, a transformer model that is pretrained on 2,000-token contexts can be well suited for processing arbitrarily short contexts, but can in some instances be architecturally incapable of meaningfully processing context windows longer than the 2,000-token training context length. In some instances, a computational cost (e.g., energy cost, etc.) of retraining such a model to facilitate a longer context window can be prohibitively expensive (e.g., hundreds of millions of dollars). In some instances, a sequence processing model pretrained on a 2,000-token context can be architecturally capable of being further trained (e.g., “fine-tuned”) to facilitate the use of longer-context-window data (e.g., 65,000 tokens, etc.), but such additional training may require a significant amount of data and computational cost. Additionally, a context window of such a model will still be limited by a size of the training context window used during the additional training process.

[0061] Advantageously, systems and methods of the present disclosure can enable the use of arbitrarily long context windows without any need for model retraining. In thismanner, for instance, systems and methods of the present disclosure can be associated with a lower computational cost and lower energy cost than retraining-based methods (e.g., pretraining a new model, fine-tuning a small-context-window model on longer-context data, etc.). For example, in some instances, training a sequence processing model to enable a larger context window can cost millions or hundreds of millions of dollars. In contrast, provided systems and methods can be efficiently combined with already-existing models to provided extended context windows at a low computational and energy cost.

[0062] Additionally, systems and methods of the present disclosure may in some instances be associated with a lower computational cost of inference compared to already- existing long-context-window models, even when retraining is not required. For example, processing a context of n tokens with a sequence generation model having an embedding dimension of d can in some instances be associated with a computational cost proportional to ri2d. Additionally, high-quality sequence generation can in some instances depend on a high- computational-cost sequence generation model (e.g., having a large number of parameters). Advantageously, provided systems and methods can process a long context using a lower- computational-cost sequence embedding model, and can provide a shorter condensed context for processing by a higher-computational -cost sequence generation model. In this manner, for instance, provided systems and methods can process a context of a given length at a lower cost and lower latency than prior systems and methods, even in instances where the prior systems and methods may not require retraining.

[0063] Similarly, systems and methods of the present disclosure can be associated with a lower energy cost and lower latency than iterative summarization methods for providing a condensed context. For example, iterative summarization can in some instances require repeatedly calling a high-computational -cost sequence generation model (e.g., large language model having hundreds of billions of parameters, etc.). For example, a high-cost model may be called repeatedly not only to summarize a plurality of semantic chunks of a long context, but also to summarize a plurality of summaries. In contrast, systems and methods of the present disclosure permit the use of a lightweight machine-learned sequence embedding model, along with a lightweight clustering system, to populate a finite context window at a low computational cost. Additionally, provided systems and methods require each semantic chunk to be embedded only once. Additional data (e.g., an additional book or other document) can be added to an arbitrarily long context of the present disclosure without re-embedding any existing data (e.g., by performing a lightweight reclustering operation).

[0064] Additionally, systems and methods of the present disclosure can be efficiently combined with other systems and methods for extending a context window length. For example, provided systems and methods can efficiently embed and cluster semantic chunks of arbitrarily large contexts (e.g. databases or corpora having millions, hundreds of millions, billions of tokens, etc.) and select chunks for inclusion into finite context windows of any size (e.g. thousands of tokens, hundreds of thousands of tokens, etc.). Similarly, provided semantic clustering methods can be combined with summarization methods to select a plurality of summaries for a finite context window at a lower energy cost compared to prior methods.

[0065] A technical effect of example implementations of the present disclosure is increased energy efficiency in performing operations using machine-learned models, thereby improving the functioning of computers implementing such models. For instance, example implementations can provide for more energy-efficient runtime execution or inference. In some scenarios, increased energy efficiency can provide for less energy to be used to perform a given task (e.g., less energy expended to maintain the model in memory, less energy expended to perform calculations within the model, etc.). In some scenarios, increased energy efficiency can provide for more task(s) to be completed for a given energy budget (e.g., a larger quantity of tasks, more complex tasks, the same task but with more accuracy or precision, etc.).

[0066] In another example aspect, example implementations can provide for more energy-efficient training operations or model updates. In some scenarios, increased energy efficiency can provide for less energy to be used to perform a given number of update iterations (e.g., less energy expended to maintain the model in memory, less energy expended to perform calculations within the model, such as computing gradients, backpropagating a loss, etc.). In some scenarios, increased energy efficiency can provide for more update iterations to be completed for a given energy budget (e.g., a larger quantity of iterations, etc.). In some scenarios, greater expressivity afforded by model architectures and training techniques of the present disclosure can provide for a given level of functionality to be obtained in fewer training iterations, thereby expending a smaller energy budget. In some scenarios, greater expressivity afforded by model architectures and training techniques of the present disclosure can provide for an extended level of functionality to be obtained in a given number of training iterations, thereby more efficiently using a given energy budget.

[0067] In this manner, for instance, the improved energy efficiency of example implementations of the present disclosure can reduce an amount of pollution or other wasteassociated with implementing machine-learned models and systems, thereby advancing the field of machine-learning and artificial intelligence as a whole. The amount of pollution can be reduced in toto (e.g., an absolute magnitude thereof) or on a normalized basis (e.g., energy per task, per model size, etc.). For example, an amount of CO2 released (e.g., by a power source) in association with training and execution of machine-learned models can be reduced by implementing more energy-efficient training or inference operations. An amount of heat pollution in an environment (e.g., by the processors / storage locations) can be reduced by implementing more energy-efficient training or inference operations.

[0068] Various example implementations are described herein with respect to the accompanying Figures.Example Systems

[0069] Figure l is a block diagram of an example system according to the present disclosure, wherein semantic clustering can enable the use of an arbitrarily long input context with a machine-learned model having a finite context window. A long-form context 102 can be obtained by a computing system 104, which can split the long-form context 102 into a plurality of context chunks 106. A lightweight sequence embedding model 108 can generate semantic chunk embeddings 110 based on the context chunks 106. A clustering system 112 can cluster the semantic chunk embeddings 110 into a plurality of semantic clusters 114 comprising one or more semantic chunk embeddings 110 per cluster. In some instances, the semantic clusters 114 can be provided as input to one or more machine-learned models, such as a sequence generation model 116, which can generate synthesized insights 118 based on the semantic clusters 114. Based on the semantic clusters 114, the computing system 104 can select one or more context chunks 106 or synthesized insights 118 to use as selected context 120. The selected context 120 can be provided as input to a sequence generation model 116, which can generate a generated output 122 based on the selected context 120.

[0070] Figure 1 depicts a long-form context 102. The long-form context 102 can generally include or otherwise represent various types of data. In some instances, the long- form context 102 can include various types of sequence data (e.g., image data, audio data, text data, etc.). The long-form context 102 can include one type or many types of data. In some instances, the long-form context 102 can include multi-modal data (e.g., sequence data having multiple modes such as video and audio; sequence data combined with non-sequence data or metadata; etc.). In some instances, the long-form context 102 can be characterized by a size that is larger than a context window of a machine-learned model of interest (e.g., sequence generation model 116 as described below). In some instances, the long-formcontext 102 can be stored or located in one location or distributed across a plurality of locations (e.g., stored in one file, document, memory location, etc.; stored in a plurality of files, documents, memory locations, etc.; stored on one computing device or a plurality of computing devices; etc.). In some instances, long-form context 102 can comprise one or more corpora (e.g., datasets, databases, etc.), wherein each corpus of the one or more corpora can comprise one or more sequences (e.g., books, documents, audio files, images, etc.). Long- form context 102 can include data obtained in one way or a plurality of ways, such as received from a user input; retrieved from one or more computer-readable media; retrieved via an internet search; received from another computing device; etc.

[0071] Figure 1 depicts a computing system 104. The computing system 104 can include, for example, one or more computing devices. The computing system 104 can be located on a single computing system or distributed across multiple computing systems. In some instances, a computing system 104 can correspond to a computing system described with respect to Figures 4 through 13 (e.g., server computing system 60, etc.).

[0072] Figure 1 depicts context chunks 106. The context chunks 106 can generally include or otherwise represent various types of data. In some instances, the context chunks 106 can include or represent various types of sequence data (e.g., image data, audio data, text data, etc.). The context chunks 106 can include one type or many types of data. In some instances, the context chunks 106 can include multi-modal data (e.g., sequence data having multiple modes such as video and audio; sequence data combined with non-sequence data or metadata; etc.). In some instances, context chunks can be, comprise, or otherwise correspond to subsequences of the long-form context 102. In some instances, the context chunks 106 can be characterized by a size that is smaller than a context window of a machine-learned model of interest (e.g., sequence generation model 116 as described below). In some instances, the context chunks 106 can be characterized by a size that is much smaller than a context window of a model of interest, such that the context window is capable of receiving a plurality of context chunks 106.

[0073] Context chunks 106 can be generated in various ways. A computing system 104 can be configured to generate context chunks 106 using one mechanism or multiple mechanisms. In some instances, context chunks 106 can be generated by splitting a long-form context 102, which can be accomplished in various ways.

[0074] Splitting a long-form context 102 can include, for example, splitting a long- form context 102 based on one or more pre-existing boundaries associated with, for example, a computing system 104, dataset, or computer-readable media. For example, long-formcontext 102 can include a plurality of files or other data types, and splitting the long-form context 102 can include splitting by file, etc.

[0075] Splitting a long-form context 102 can include, for example, splitting by one or more sequence boundaries. A sequence boundary can include, for example, a natural boundary or a metadata-defined boundary. A natural boundary can include, for example, a boundary that can be identified based on sequence data alone, without reference to metadata not included in the sequence. As an illustrative example, a natural boundary can include a grammatically defined boundary between natural language components (e.g., paragraphs, sentences, etc.). Other natural sequence boundaries can include, for example, boundaries between songs; periods of relative silence in an audio sequence; boundaries between objects depicted in an image; boundaries between acts or scenes in a work of fiction; conversational turns; etc. A metadata-defined boundary can include, for example, a boundary that can be defined based on metadata associated with part of a sequence (e.g., a file component defining a paragraph in a document file, such as a Visual Basic Paragraph object or ListParagraph object; a chapter number associated with textual sequence data located in a book or PDF file; a track number or chapter boundary associated with an audio file or video file; etc.). In some instances, a metadata-defined boundary can include, for example, a document boundary (e.g., page boundary, chapter boundary, section boundary, etc.) associated with a document (e.g., electronic document, physical book, etc.) from which part of the long-form context 102 was obtained.

[0076] Splitting a long-form context 102 can include, for example, splitting based on one or more delimiters. A delimiter can include, for example, a token or plurality of tokens of the long-form context 102 (e.g., new-line character in a textual sequence, period of silence in an audio sequence, predefined pixel sequence, etc.) For example, in some instances a long- form context 102 can include structured data (e.g., XML data, structured few-shot prompts or chain-of-thought prompts configured as input for a machine-learned model, etc.) and a delimiter can be a sequence associated with a structure associated with the structured data (e.g., XML tag, structured few-shot prompt tag such as “Problem:”, “First step:”, etc.).

[0077] Splitting a long-form context 102 can include, for example, splitting based on one or more chunk length thresholds, such as a minimum or maximum chunk length. In some instances, a long-form context 102 can be split based on a fixed chunk length without regard to sequence content. In other examples, splitting by length threshold can be combined with other splitting mechanisms. For example, a maximum chunk length can be defined (e.g., a maximum length associated with a context window of the lightweight sequence embeddingmodel, etc.); a long-form context 102 can be split in one or more ways described herein (e.g., split by paragraph); and any context chunk 106 longer than the maximum length can be split in various ways (e.g. split in half, split at a sentence boundary, etc.).

[0078] In some instances, splitting a long-form context 102 into context chunks 106 can include using a machine-learned model. For example, in some instances a machine- learned model can be used to identify one or more natural boundaries of a sequence (e.g., a voice-recognition model to identify conversational turns in an audio sequence; an image processing model to identify objects in an image; etc.). In some instances, a machine-learned model can be used to determine metadata associated with the sequence (e.g., whether an audio sequence contains music, speech, sounds of nature.; chapter number, page number, or figure number associated with a textual sequence or subsequence; etc.), and the long-form context 102 can be split based on the metadata. In some instances, a machine-learned model can be used to identify a semantic similarity or dissimilarity between subsequences of the long-form context 102, and the context chunks 106 can be defined in part by the similarity or dissimilarity (e.g., paragraphs with highly dissimilar sentences can be split; two highly similar paragraphs can be merged into a single context chunk 106; etc.)

[0079] Figure 1 depicts a lightweight sequence embedding model 108. The lightweight sequence embedding model 108 can be or include various different types of machine-learned model architectures. In some instances, the lightweight sequence embedding model 108 can be or include a sequence processing model that is configured to generate one or more output(s) based on one or more input(s) in a sequence. For example, the lightweight sequence embedding model 108 can be or include a sequence processing model that is configured to generate one or more semantic embeddings based on various types of sequence data (e.g., image data, audio data, text data, multimodal data, etc.). In some instances, the lightweight sequence embedding model 108 can leverage an attention mechanism such as self-attention. For example, the lightweight sequence embedding model 108 can be or include a multi-headed self-attention model (e.g. an encoder-only, encoder-decoder, or decoder-only transformer language model such as PaLM, etc.). In some instances, the lightweight sequence embedding model 108 can be or include a word2vec model or similar model configured for multi-word sequences (e.g., doc2vec, etc.). In some instances, the lightweight sequence embedding model 108 can be or include an embedding model configured for use with a token retrieval mechanism (e.g., XTR, etc.).

[0080] In some instances, the lightweight sequence embedding model 108 can be or include a model configured to receive a context chunk 106 as input and generate one or moresemantic embeddings based on the context chunk 106. In some instances, lightweight sequence embedding model 108 can have a token window that is sufficiently long to generate a single semantic embedding associated with an entire context chunk 106. In such instances, the generated semantic embedding can be a semantic chunk embedding 110. In other instances, the lightweight sequence embedding model 108 can be configured to generate more than one semantic embedding for a given context chunk 106 (e.g., one semantic embedding per token; one semantic embedding for subsequence of a fixed maximum length, wherein a length of the context chunk 106 is greater than the maximum length; etc.). In such instances, generating a semantic chunk embedding 110 can include combining a plurality of semantic embeddings generated by the lightweight sequence embedding model 108 (e.g., via pooling such as average pooling; machine-learned combination; concatenation; etc.) to generate a single semantic chunk embedding 110 associated with a context chunk 106.

[0081] In some instances, the lightweight sequence embedding model 108 can be associated with a lower computational cost (e.g., energy cost, latency, etc.) relative to the sequence generation model 116. In some instances, the lightweight sequence embedding model 108 can be characterized by a first number of parameters (e.g., 100 million, 500 million, 1 billion, 5 billion, etc.) that is smaller than a second number of parameters associated with the sequence generation model 116. In some instances, the second number of parameters can be much larger (e.g., five times, ten times, 30 times, 50 times, 100 times, 500 times larger, etc.) than the first number. Although Fig. 1 depicts a lightweight sequence embedding model, a sequence embedding model having a computational cost or number of parameters that is the same as or larger than a computational cost or number of parameters of the sequence generation model 116 can be used in place of the lightweight sequence embedding model 108 without going outside the scope of the present disclosure.

[0082] Figure 1 depicts semantic chunk embeddings 110. A semantic chunk embedding 110 can be, for example, a machine-learned semantic embedding associated with a context chunk 106. A semantic chunk embedding 110 can generally be, include, or be represented by any computer-readable data type (e.g., vector, matrix, or tensor data; binary data; numerical data; etc.).

[0083] Figure 1 depicts a clustering system 112. The clustering system 112 can include, for example, one or more computing devices. The clustering system 112 can be located on a single computing system or distributed across multiple computing systems. In some instances, the clustering system 112 can be, comprise, be comprised by, implement, or be implemented by a computing system 104. In some instances, a clustering system 112 cancorrespond to a computing system described with respect to Figures 4 through 13 (e.g., server computing system 60, etc.).

[0084] Figure 1 depicts semantic clusters 114. A semantic cluster 114 can include, for example, data indicative of a plurality of semantic chunk embeddings 110 or context chunks 106. In some instances, a semantic cluster 114 can include, for example, one or more subclusters having one or more context chunks 106 per subcluster. The semantic clusters 114 can be generated using any appropriate clustering algorithm (e.g., centroid-based clustering such as k-means, distribution-based clustering such as g-means, connectivity-based clustering such as hierarchical agglomerative clustering, dimensionality-reduced clustering such as spectral clustering, density-based clustering, biclustering, hierarchical divisive clustering, affinity propagation, etc.). In some instances, generating semantic clusters 114 can include determining a target number of clusters (e.g., based on a size of a finite context window of a sequence generation model 116). In some instances, a clustering algorithm (e.g., k-means, hierarchical agglomerative clustering, hierarchical divisive clustering, etc.) can be configured to generate a number of clusters equal to or approximately equal to the target number. In some instances, a semantic chunk embedding 110 associated with a first semantic cluster 114 can be more similar, according to one or more measures of similarity, to the first semantic cluster 114 than to any other semantic cluster 114. For example, a semantic chunk embedding 110 associated with a first semantic cluster can in some instances be closer, according to a measure of semantic distance (e.g., cosine distance associated with a vector-based semantic embedding, etc.), to a centroid of the first semantic cluster 114 than to any centroid of any other semantic cluster 114. An example implementation for generating semantic clusters 114 is further described below with respect to Figure 2.

[0085] Figure 1 depicts a sequence generation model 116. The sequence generation model 116 can be or include various different types of machine-learned model architectures. The sequence generation model 116 can be or include a sequence processing model that is configured to generate one or more output(s) in a sequence based on one or more input(s). In some instances, the input(s) can include sequence data (e.g., text, image, audio, multimodal, etc.). In some instances, the sequence generation model 116 can be or include a model configured to generate one or more output(s) in a sequence based on one or more semantic embeddings (e.g., semantic embeddings generated by the lightweight sequence embedding model 108). For example, in some instances, a lightweight sequence generation model 108 and a sequence generation model 116 can respectively comprise an encoder and a decoder configured to operate on a similar (e.g., same) semantic embedding space. In some instances,the sequence generation model 116 can leverage an attention mechanism such as selfattention. For example, the sequence generation model 116 can be a multi-headed selfattention model (e.g. an encoder-only, encoder-decoder, or decoder-only transformer language model).

[0086] Figure 1 depicts synthesized insights 118. Synthesized insights 118 can generally include or otherwise represent various types of data. The synthesized insights 118 can include one type or many types of data. In some instances, the synthesized insights 118 can include multi-modal data (e.g., multiple types of sequence data such as video and audio; sequence and non-sequence data or metadata; etc.). In some instances, the synthesized insights 118 can be characterized by a size that is smaller than a context window of a machine-learned model of interest (e.g., sequence generation model 116). In some instances, the synthesized insights 118 can be characterized by a size that is much smaller than a context window of a model of interest, such that the context window is capable of receiving a plurality of synthesized insights 118 or context chunks 106. Synthesized insights 118 can include, for example, sequence data generated based on one or more context chunks 106 or semantic chunk embeddings 110 (e.g., summaries, suggested questions, main ideas, topic descriptions, etc.). Synthesized insights 118 can include, for example, non-sequence data (e.g., metadata such as a sequence type, a numerical or categorical topic identifier, a number of times a particular token appears in a sequence, etc.). Although Figure 1 depicts a sequence generation model 116 generating synthesized insights 118, synthesized insights 118 can be generated without using a machine-learned model (e.g., using a computing system 104) or using a machine-learned model other than a sequence generation model 116 without going outside the scope of the present disclosure.

[0087] Figure 1 depicts selected context 120. The selected context 120 can generally include or otherwise represent various types of data. In some instances, the selected context 120 can include or represent various types of sequence data (e.g., image data, audio data, text data, etc.). In some instances, the selected context 102 can include or represent various types of non-sequence data (e.g., metadata, synthesized insights 118, etc.) In some instances, the selected context 120 can include multi-modal data (e.g., sequence data of multiple modes such as video and audio; sequence data combined with non-sequence data; etc.). In some instances, the selected context 120 can include or represent semantic embedding data (e.g., when a sequence generation model 116 comprises a model configured to generate a sequence output from a semantic embedding input). In some instances, selected context 120 cancomprise one or more context chunks 106. In some instances, selected context 120 can comprise one or more synthesized insights 118.

[0088] Figure 1 depicts the computing system 104 determining the selected context 120 based on one or more semantic clusters 114 or synthesized insights 118. Determining selected context 120 can include, for example, deterministic or probabilistic selection based on one or more priorities. In some instances, selecting based on the one or more priorities can include, for example, prioritizing context chunks 106 associated with semantic chunk embeddings 110 near a centroid of a semantic cluster 114 according to a measure of semantic distance (e.g., cosine distance, etc.). A centroid of a semantic cluster 114 can be defined in various ways, such as by averaging (e.g. arithmetic average, geometric average, etc.) a plurality of semantic chunk embeddings 110 associated with the semantic cluster 114. In some instances, selecting based on the one or more priorities can include, for example, prioritizing context chunks 106 associated with semantic chunk embeddings 110 near a centroid of a subcluster of a hierarchically defined semantic cluster 114 according to a measure of semantic distance.

[0089] In some instances, selecting based on one or more priorities can include prioritizing context chunks 106 associated with semantic clusters 114 having a large number of context chunks 106. In some instances, selecting based on one or more priorities can include prioritizing context chunks 106 associated with a semantic cluster 114 that is currently unrepresented in the selected context 120. In some instances, selecting based on one or more priorities can include prioritizing context chunks 106 associated with a hierarchically defined subcluster of a semantic cluster 114 that is currently unrepresented in the selected context. In some instances, one or more priorities can be determined based on a user input. For example, in some instances, selecting based on one or more priorities can include prioritizing semantic chunk embeddings 110 or semantic cluster 114 centroids near a semantic embedding determined based on a user input. In some instances, selecting based on one or more priorities can include prioritizing context chunks 106 associated with a high informational content (e.g., based on a mathematical definition of information entropy) relative to context chunks 106 that have already been selected for the selected context 120.

[0090] In some instances, selecting based on one or more priorities can comprise a hierarchical selection mechanism such as a decision tree. In some instances, selecting based on one or more priorities can comprise a numerical selection mechanism such as a weighted combination of priorities. For example, in some instances, a selection mechanism can comprise assigning a respective priority score to each context chunk 106, wherein a priorityscore can comprise, for example, a weighted average of a plurality of priority subscores (e.g., distance from nearest cluster or subcluster centroid, number of context chunks 10 in cluster, etc.).

[0091] In some instances, determining a selected context 120 can include generating a context sequence based directly on a semantic embedding. For example, in some instances, a centroid of a semantic cluster 114 or a subcluster of a semantic cluster 114 can be determined. In some instances, the centroid itself can be decoded without respect to any particular context chunk 106 of the semantic cluster 114. For example, in some instances, a machine-learned sequence generation model can generate or otherwise decode a context sequence based on a location of a centroid in a semantic embedding space. In some instances, determining a selected context 120 can include decoding a context sequence based on a semantic embedding other than the centroid. For example, determining a selected context can include decoding a context sequence based on a geometric transformation of a centroid or a geometric transformation of a semantic chunk embedding 110. For example, a geometric transformation can include a transformation configured to correspond to a particular semantic relationship (e.g., opposites, hierarchical relationships such as is-a or has-a relationships, analogical relationships such as king : queen :: prince : princess, etc.).

[0092] In some instances, a selected context 120 can include one or more prompt sequences (e.g., “I’m going to give you some representative paragraphs from a book. Please summarize the book based on the following paragraphs:”, etc.). In some instances, a prompt sequence can include one or more user inputs (e.g., a question or instruction received from a user, etc.).

[0093] Figure 1 depicts generated output 122. The generated output 122 can generally include or otherwise represent various types of data. In some instances, the generated output 122 can include or represent various types of sequence data (e.g., image data, audio data, text data, multimodal data, etc.) and non-sequence data or metadata. The generated output 122 can include one or more data types that are the same as or different from a data type of the long-form context 102, context chunks 106, synthesized insights 118, etc.

[0094] Figures 2A through 2C are illustrations of an example hierarchical clustering algorithm, wherein a clustering system 112 can group semantic chunk embeddings 110 into semantic clusters 114. Based on a plurality of unclustered semantic chunk embeddings 210 in a semantic embedding space 202, a clustering system 112 can perform a first clustering to cluster semantic chunk embeddings 210 into a plurality of first clusters 214a-f comprising one or more clustered semantic chunk embeddings 212a-f in each first cluster. Based on thefirst clusters 214a-f, the clustering system 112 can perform a second clustering to create a plurality of second clusters 216 comprising one or more first clusters in each second cluster. In some instances, the first clusters or second clusters 216 can be semantic clusters 114.

[0095] Figures 2A through 2C depict a semantic embedding space 202. A semantic embedding space 202 can include, for example, a space of possible semantic embeddings (e.g., semantic chunk embeddings 110) that are capable of being generated by a machine- learned model (e.g., lightweight sequence embedding model 108) based on a plurality of possible input values (e.g., possible values for a context chunk 106). A semantic embedding space 202 can include or represent, for example, any data type capable of representing a semantic embedding (e.g., semantic chunk embedding 110). In some instances, a semantic embedding space can be, for example, an ^-dimensional vector space, where n can be an integer. Although the illustration of Figure 2 is rendered in two dimensions to improve readability, it will be understood that n can in some instances be greater (e.g., much greater) than two.

[0096] Figure 2A depicts a plurality of unclustered semantic chunk embeddings 210. An unclustered semantic chunk embedding 210 can be, for example, a semantic chunk embedding 110 that has not been assigned to a cluster. When a clustering system 112 first obtains semantic chunk embeddings 110, all semantic chunk embeddings 110 can in some instances be unclustered semantic chunk embeddings 210.

[0097] Figures 2A-C depict a plurality of clustered semantic chunk embeddings 212a- f. A clustered semantic chunk embedding 212 can be, for example, a semantic chunk embedding 110 that has been assigned to a cluster.

[0098] Figures 2B-C depict a plurality of first clusters 214a-f. In some instances, a first cluster 214 can be, comprise, be comprised by, or share one or more properties with a semantic cluster 114.

[0099] Figure 2C depicts a plurality of second clusters 216. In some instances, a second cluster 216 can be, comprise, be comprised by, or share one or more properties with a semantic cluster 114.

[0100] A clustering system 112 can, for example, assign one or more un clustered semantic chunk embeddings 210 to one or more initial clusters. For example, Figure 2A depicts four clustered semantic chunk embeddings 212a-d that have been assigned to four respective first clusters 214, wherein each cluster contains only one clustered semantic chunk embedding 212 at first. In some instances, the clustering system 112 can determine one or more initial clustered semantic chunk embeddings 212a-d by random sampling. In someinstances, the clustering system 112 can determine one or more initial clustered semantic chunk embeddings in another manner (e.g., based on a plurality of respective distances between a first clustered semantic chunk embedding 212a and a plurality of respective unclustered semantic chunk embeddings 210, etc.).

[0101] Based on the initial clustered semantic chunk embeddings 212a-d, the clustering system 112 can assign additional unclustered semantic chunk embeddings 210 to one or more clusters 214. In some instances, the assignments can be based on a plurality of respective distances (e.g., semantic distances) between an unclustered semantic chunk embedding 210 and a plurality of respective centroids of a plurality of respective first clusters 214a-d. In some instances, a location in semantic space 202 of a respective centroid can be updated after each new clustered semantic chunk embedding 212 is added to the first cluster 214. In some instances, one or more clustered semantic chunk embeddings 212 can be removed from a cluster or reassigned to another cluster. In some instances, such a reassignment can be based on a plurality of respective distances from a clustered semantic chunk embedding to a plurality of respective updated centroids of a plurality of first clusters 214.

[0102] In some instances, a final number of first clusters 214 can be predetermined, and a number of initial first clusters 214a-d can be equal to the final number of first clusters 214 (e.g., via A means clustering). In other instances, a number of first clusters 214 can be dynamically determined. For example, in some instances, a number of first clusters 214 can be dynamically determined based at least in part on a target distribution of one or more first clusters 214. For example, in some instances, g-means clustering can be performed, wherein a number of first clusters is dynamically chosen such that a distribution of one or more first clusters 214 is approximately Gaussian. In some instances, a number of first clusters 214 can be dynamically determined based on other factors, such as one or more cluster density thresholds. Figures 2A-B depict a dynamically determined number of first clusters 214, wherein 2 A depicts a clustering system 112 initially selecting four clustered semantic chunk embeddings 212a-d for four initial first clusters 214a-d (e.g., by random sampling), while Figure 2B depicts a final number of six first clusters 214a-f.

[0103] Figure 2C depicts a plurality of second clusters 216 determined by the clustering system 112 based on the plurality of first clusters 214a-f. Determining a plurality of second clusters 216 can include, for example, hierarchical agglomerative clustering (e.g., parhac, etc.). For example, in some instances two or more first clusters 214a-f can be merged based on one or more similarities between the two or more first clusters 214a-f. Thesimilarities can be based on, for example, one or more distances in the semantic embedding space 202 (e.g., shortest minimum distance or shortest maximum distance between pairs of clustered semantic chunk embeddings 212a, 212b; shortest distance between centroids of first clusters 214; average, median, or weighted average distance between pairs of clustered semantic chunk embeddings 212a, 212b; minimum increase in variance; minimum increase in sum of squares; etc.). In some instances, a final number of second clusters can be determined based at least in part on a size of a finite context window of the sequence generation model 116. For example, a final number of second clusters can be based at least in part on a ratio between a finite context window size and a size of one or more context chunks 106. In such instances, a second clustering can continue to merge clusters until a target number of second clusters is reached.

[0104] Although Figures 2A-C include a variety of example details for the purposes of illustration, alternative implementations are possible without going outside the scope of the present disclosure. For example, a number of clusters can be larger and smaller than a number depicted in Figures 2A-C. Distances and boundaries depicted in Figures 2A-C are not necessarily to scale. A variety of different clustering methods can be used to generate semantic clusters 114, whether depicted in Figures 2A-C or not (e.g., centroid-based clustering such as k-means, distribution-based clustering such as g-means, connectivity-based clustering such as hierarchical agglomerative clustering, dimensionality -reduced clustering such as spectral clustering, density-based clustering, biclustering, hierarchical divisive clustering, affinity propagation, etc.). Although Figures 2A-C depict a first clustering and a second clustering, it will be appreciated that a second clustering can be omitted, or a third clustering and so on can be added. For example, in some instances, hierarchical agglomerative clustering can be used alone, wherein each unclustered semantic chunk embedding 210 is initially treated as a one-chunk “cluster” to be merged with other clusters until a target number of clusters is achieved.Example Methods

[0105] Figure 3 depicts a flowchart diagram of an example method for semantic clustering according to example embodiments of the present disclosure. Although Figure 3 depicts steps performed in a particular order for purposes of illustration and discussion, the methods of the present disclosure are not limited to the particularly illustrated order or arrangement. The various steps of example method 300 can be omitted, rearranged,combined, and / or adapted in various ways without deviating from the scope of the present disclosure.

[0106] At 302, example method 300 can include obtaining, by one or more computing devices, a context sequence. In some instances, a computing device can be, comprise, or be comprised by a computing system 104. In some instances, a context sequence can be, comprise, or be comprised by a long-form context 102. In some instances, example method 300 can include, at 302, using one or more systems or performing one or more activities described with respect to Figure 1.

[0107] At 304, example method 300 can include determining, by the one or more computing devices, a plurality of subsequences of the context sequence. In some instances, a subsequence can be, comprise, or be comprised by a context chunk 106. In some instances, determining the plurality of subsequences of the context sequence can include splitting the context based at least in part on a subsequence length threshold. In some instances, determining the plurality of subsequences of the context sequence can include splitting the context based at least in part on one or more delimiters. In some instances, at least one of the one or more delimiters can include one or more numerical, alphabetical, textual, or whitespace characters. In some instances, at least one of the one or more delimiters can include one or more document boundaries. In some instances, determining the plurality of subsequences of the context sequence can include using a machine-learned model. In some instances, example method 300 can include, at 304, using one or more systems or performing one or more activities described with respect to Figure 1.

[0108] At 306, example method 300 can include determining, by the one or more computing devices using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence. In some instances, a respective semantic embedding can be, comprise, or be comprised by a semantic chunk embedding 110. In some instances, the first machine- learned model can be configured to receive, as input, a sequence comprising multiple tokens; and generate a single semantic embedding based on at least two different tokens of the multiple tokens. In some instances, the first machine-learned model can be configured to generate one semantic embedding for each input token it receives. In the example method, determining a semantic embedding associated with a subsequence of the context sequence can include combining a plurality of semantic embeddings generated by the first machine- learned model based on the subsequence. In some instances, example method 300 caninclude, at 306, using one or more systems or performing one or more activities described with respect to Figure 1.

[0109] At 308, example method 300 can include determining, by the one or more computing devices, a plurality of semantic clusters, wherein each semantic cluster of the plurality of semantic clusters comprises one or more semantic embeddings of the plurality of semantic embeddings. In some instances, a semantic cluster can be, comprise, or be comprised by a semantic cluster 114. In some instances, determining the plurality of semantic clusters can include hierarchical clustering. In some instances, example method 300 can include, at 308, using one or more systems or performing one or more activities described with respect to Figures 1 or 2.

[0110] At 310, example method 300 can include selecting, by the one or more computing devices based at least in part on the plurality of semantic clusters, one or more subsequences of the plurality of subsequences of the context sequence. In some instances, a selected subsequence of the context sequence can be, comprise, or be comprised by a selected context 120. In some instances, selecting the one or more subsequences can include selecting based on one or more priorities. In some instances, selecting based on the one or more priorities can include prioritizing clusters associated with a larger number of semantic embeddings. In some instances, selecting based on the one or more priorities can include prioritizing semantic embeddings that are near a center of a semantic cluster. In some instances, the center of the semantic cluster can be determined by averaging a plurality of semantic embeddings of the semantic cluster. In some instances, the one or more priorities can include a priority determined based on a user input. In some instances, selecting based on the one or more priorities can include prioritizing semantic embeddings that are near a semantic embedding associated with a user input. In some instances, selecting based on the one or more priorities can include prioritizing semantic embeddings from a cluster having a centroid near a semantic embedding associated with a user input. In some instances, at least one semantic cluster of the plurality of semantic clusters can include a plurality of subclusters. In some instances, selecting based on the one or more priorities can include prioritizing a selection of at least one subsequence from each of two or more subclusters of the plurality of subclusters over a selection of a second subsequence from any subcluster of the plurality of subclusters. In some instances, selecting based on the one or more priorities can include prioritizing semantic embeddings near a centroid of a subcluster of the plurality of subclusters. In some instances, selecting based on the one or more priorities can include prioritizing a selection of at least one subsequence from each semantic cluster of the pluralityof semantic clusters over a selection of a second subsequence from any semantic cluster of the plurality of semantic clusters. In some instances, example method 300 can include, at 310, using one or more systems or performing one or more activities described with respect to Figure 1.

[0111] As one example, an example selection algorithm can start with a largest semantic cluster of the plurality of semantic clusters and select a subsequence having a semantic embedding closest to a center of the largest semantic cluster. Next, an example selection algorithm can select from the second largest cluster, then the third largest cluster, and so on until either: (a) the context window size limit is met, or (b) at least one respective subsequence has been chosen from each respective semantic cluster of the plurality of semantic clusters. If the context window has room remaining after each semantic cluster is represented in the context window, then a second cluster can be selected from the largest cluster. In some instances, the second cluster can have an embedding that is second closest to a center of the largest semantic cluster, or closest to a center of a hierarchical subcluster of the largest semantic cluster. After each respective semantic cluster is represented in a context window, a respective number of additional subsequences added from a respective cluster (e.g., largest cluster, second largest cluster, etc.) can be based on (e.g., approximately proportional to) a size of the respective cluster.

[0112] In another example, an example selection algorithm can start with a semantic cluster having a center that is closest to a semantic embedding of a user input. The example algorithm can select a subsequence from the cluster that is either, for example: (a) closest to a center of the cluster, or (b) closest to the semantic embedding of the user input. Next, the example algorithm can select from the cluster that is second closest to the semantic embedding of the user input, and so on until all clusters are represented. After each respective semantic cluster or subcluster is represented in a context window, a respective number of additional subsequences added from a respective cluster or subcluster (e.g., closest cluster to a user input, second closest cluster, etc.) can be based on (e.g., inversely proportional to) a distance between a center of the respective cluster and the semantic embedding of the user input.

[0113] At 312, example method 300 can include adding, by the one or more computing devices, data associated with the one or more subsequences to a context window of a second machine-learned sequence processing model. In some instances, a second machine-learned sequence processing model can be, comprise, or be comprised by a sequence generation model 116. In some instances, a second machine-learned model can bethe first machine-learned model, or a model other than the first machine-learned model. In some instances, the data associated with the one or more subsequences can include at least one of the one or more subsequences. In some instances, the data associated with the one or more subsequences can include metadata associated with at least one of the one or more subsequences. In some instances, the data associated with the one or more subsequences can include a machine-generated summary of at least one of the one or more subsequences. In some instances, the data associated with the one or more subsequences can include one or more tokens generated by the one or more computing devices based at least in part on a semantic embedding associated with at least one of the one or more subsequences. In some instances, the data associated with the one or more subsequences can include one or more respective semantic embeddings of the plurality of respective semantic embeddings. In some instances, the data associated with one or more subsequences can include one or more tokens generated by the one or more computing devices based at least in part on a geometric transformation of a semantic embedding associated with at least one of the one or more subsequences. In some instances, example method 300 can include, at 312, using one or more systems or performing one or more activities described with respect to Figure 1.

[0114] In some instances, example method 300 can include, at 312, adding a prompt to the context window of the second machine-learned model. In some instances, the prompt can include one or more instructions or explanations.

[0115] In some instances, the first machine-learned model can be characterized by a first number of parameters; the second machine-learned model can be characterized by a second number of parameters; and the second number can be at least five times larger than the first number. In some instances, the second number can be at least 10 times larger than the first number. In some instances, the second number can be at least 30 times larger than the first number.

[0116] At 314, example method 300 can include generating, by the one or more computing devices using the second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters. In some instances, an output sequence can be, comprise, or be comprised by a generated output 122. In some instances, the output sequence can be generated based at least in part on the context window. In some instances, example method 300 can include, at 314, using one or more systems or performing one or more activities described with respect to Figure 1.

[0117] Figure 4 depicts a flowchart of a method 400 for training one or more machine-learned models according to aspects of the present disclosure. For instance, anexample machine-learned model can include a lightweight sequence embedding model 108 or sequence generation model 116.

[0118] One or more portion(s) of example method 400 can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of example method 400 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 400 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 4 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 4 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 400 can be performed additionally, or alternatively, by other systems.

[0119] At 402, example method 400 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 400 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 / learning). Example data types for the training instance and various tasks associated therewith are described throughout the present disclosure.

[0120] At 404, example method 400 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.

[0121] At 406, example method 400 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), predictedor 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).

[0122] At 408, example method 400 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 400 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0123] In some implementations, example method 400 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.).

[0124] In some implementations, example method 400 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 400 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 400 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 reinforcementlearning. Reinforcement learning can be based on user feedback on model performance during use.Example Machine-Learned Models

[0125] Figure 5 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.

[0126] 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.

[0127] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recurrent neural networks, convolutional neural networks (CNNs), diffusion models, generative-adversarial networks, or other forms of neural networks. Example neural networks can be deep neural networks. Some example machine-learned models can leverage an attention mechanism such as self-attention. For example, some example machine-learned models can include multiheaded self-attention models.

[0128] 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).

[0129] 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 types of data.

[0130] 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 executeddirectly 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.

[0131] 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.

[0132] 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 present disclosure are not limited to those examples noted above.Example Machine-Learned Sequence Processing Models

[0133] Figure 6 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-A , 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-A, etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7.

[0134] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences ofinformation. 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:2301.1 1325V1 (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.

[0135] 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”).

[0136] Sequence processing model(s) 4 can ingest the data from input(s) 2 and parse the data into a sequence of elements to obtain input sequence 5. For example, a portion of input data from input(s) 2 can be broken down into pieces that collectively represent the content of the portion of the input data. The pieces can provide the elements of the sequence.

[0137] 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.

[0138] 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 subwordtokenizer and detokenizer for Neural Text Processing, PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66-71 (October 31-November 4, 2018), https: / / aclanthology.org / D18-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image.

[0139] In general, arbitrary data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-M depicted in Figure 6 can be the tokens or can be the embedded representations thereof.

[0140] Prediction layer(s) 6 can predict one or more output elements 7-1, 7-2, . . . , 7- N based on the input elements. Prediction layer(s) 6 can include one or more machine-learned model architectures, such as one or more layers of learned parameters that manipulate and transform the input(s) to extract higher-order meaning from, and relationships between, input element(s) 5-1, 5-2, . . . , 5-M. In this manner, for instance, example prediction layer(s) 6 can predict new output element(s) in view of the context provided by input sequence 5.

[0141] Prediction layer(s) 6 can evaluate associations between portions of input sequence 5 and a particular output element. These associations can inform a prediction of the likelihood that a particular output follows the input context. For example, consider the textual snippet, “The carpenter’s toolbox was small and heavy. It was full of .” Example prediction layer(s) 6 can identify that “It” refers back to “toolbox” by determining a relationship between the respective embeddings. Example prediction layer(s) 6 can also link “It” to the attributes of the toolbox, such as “small” and “heavy.” Based on these associations, prediction layer(s) 6 can, for instance, assign a higher probability to the word “nails” than to the word “sawdust.”

[0142] A transformer is an example architecture that can be used in prediction layer(s) 4. See, e.g., Vaswani et al., Attention Is All You Need, ARXIV: 1706.03762v7 (Aug. 2, 2023). A transformer is an example of a machine-learned model architecture that uses an attention mechanism to compute associations between items within a context window. The context window can include a sequence that contains input sequence 5 and potentially one or more output element(s) 7-1, 7-2, . . . , 7-N. A transformer block can include one or more attention layer(s) and one or more post-attention layer(s) (e.g., feedforward layer(s), such as a multi-layer perceptron).

[0143] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as well asconvolutional 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.

[0144] 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.

[0145] 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.

[0146] 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.

[0147] 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).

[0148] 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 inputsequence 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.

[0149] Figure 7 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task). Input sequence 8 can include various data elements from different data modalities. For instance, an input modality 10-1 can include one modality of data. A data-to- sequence model 11-1 can process data from input modality 10-1 to project the data into a format compatible with input sequence 8 (e.g., one or more vectors dimensioned according to the dimensions of input sequence 8) to obtain elements 8-1, 8-2, 8-3. Another input modality 10-2 can include a different modality of data. A data-to-sequence model 11-2 can project data from input modality 10-2 into a format compatible with input sequence 8 to obtain elements 8-4, 8-5, 8- 6. Another input modality 10-3 can include yet another different modality of data. A data-to- sequence model 11-3 can project data from input modality 10-3 into a format compatible with input sequence 8 to obtain elements 8-7, 8-8, 8-9.

[0150] 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 / Jdimensions. 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.

[0151] 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.

[0152] In some implementations, the expressive power of the embedding space may not be limited to meanings associated with any particular set of tokens or other building blocks. For example, a continuous embedding space can encode a spectrum of high-order information. An individual piece of information (e.g., a token) can map to a particular point in that space: for instance, a token for the word “dog” can be projected to an embedded value that points to a particular location in the embedding space associated with canine-related information. Similarly, an image patch of an image of a dog on grass can also be projected into the embedding space. In some implementations, the projection of the image of the dog can be similar to the projection of the word “dog” while also having similarity to a projection of the word “grass,” while potentially being different from both. In some implementations, the projection of the image patch may not exactly align with any single projection of a single word. In some implementations, the projection of the image patch can align with a combination of the projections of the words “dog” and “grass.” In this manner, for instance, a high-order embedding space can encode information that can be independent of data modalities in which the information is expressed.

[0153] 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.

[0154] 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).

[0155] 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 cansubdivide 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.).

[0156] 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

[0157] Figure 8 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model development platform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0158] 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.

[0159] 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.

[0160] 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.

[0161] 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-trainedfoundational 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).

[0162] 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.

[0163] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., denoising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16.

[0164] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher- quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to finetune development model 16.

[0165] 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.

[0166] 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.

[0167] In some implementations, pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts caninclude inputs that lack exemplars. Zero-shot prompts can be within a domain within a training dataset or outside of the training domain(s).

[0168] 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.

[0169] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine- learned models. In this manner, for instance, a first model can process information about a task and output a input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0170] 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.

[0171] 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 400 described above.

[0172] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance quality where appropriate. For instance, deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error. For instance, instead of autoregressively predicting the solution to a system of equations, a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool. The tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations. Theoutput of the tool can be returned in response to the original query. In this manner, tool use can allow some example models to focus on the strengths of machine-learned models — e.g., understanding an intent in an unstructured request for a task — while augmenting the performance of the model by offloading certain tasks to a more focused tool for rote application of deterministic algorithms to a well-defined problem.

[0173] 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”).

[0174] 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.

[0175] 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.

[0176] 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.

[0177] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardware acceleration 19-2 can facilitate the configuration of the model storage and execution formats to operate optimally on different hardware resources. For instance, hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memory requirements, etc. Tools for distillation 19-3 can provide for the training of lighter-weightmodels based on the knowledge encoded in development model 16. For instance, development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12. To obtain a lightweight model for running in resource-constrained environments, a smaller model can be a “student model” that learns to imitate development model 16 as a “teacher model.” In this manner, for instance, the investment in learning the parameters and configurations of development model 16 can be efficiently transferred to a smaller model for more efficient inference.

[0178] 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.

[0179] Figure 9 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example training flow can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. FIG. 9 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. FIG. 9 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0180] 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.

[0181] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pretraining stage 22 can be implemented using one or more pre-training pipelines 17-2 over data from dataset(s) 17-1. Pre-training can be omitted, for example, if initialized model 21 is already pre-trained (e.g., development model 16 contains, is, or is based on a pre-trained foundational model or an expert model).

[0182] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Pre-trained model 23 can be the initial state if development model 16 was already pre-trained. Pre-trained model 23 can undergo fine-tuning in a fine-tuning stage 24. Fine-tuning stage 24 can be implemented using one or more fine-tuning pipelines 17-3 over data from dataset(s) 17-1. Fine-tuning can be omitted, for example, if a pre-trained model as satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0183] Fine-tuned model 29 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Fine-tuned model 29 can be the initial state if development model 16 was already fine-tuned. Fine-tuned model 29 can undergo refinement with user feedback 26. For instance, refinement with user feedback 26 can include reinforcement learning, optionally based on human feedback from human users of fine-tuned model 25. As reinforcement learning can be a form of fine-tuning, it is to be understood that fine-tuning stage 24 can subsume the stage for refining with user feedback 26. Refinement with user feedback 26 can produce a refined model 27. Refined model 27 can be output to downstream system(s) 28 for deployment or further development.

[0184] 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

[0185] Figure 10 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.

[0186] 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.

[0187] 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.

[0188] 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.

[0189] For example, model host 31 can operate on a server system that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network). Client device(s) can be end-user devices used by individuals. Clientdevice(s) can be server systems that operate client(s) 32 to provide various functionality as a service to downstream end-user devices.

[0190] 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.

[0191] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include weights or other model components that are stored on in persistent storage, temporarily cached, or loaded into high-speed memory. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include cached intermediate states of active or inactive model(s) used to accelerate inference of those models. For instance, an inference session with a particular model may generate significant amounts of computational results that can be re-used for future inference runs (e.g., using a KV cache for transformer-based models). These computational results can be saved in association with that inference session so that session can be executed more efficiently when resumed.

[0192] 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.

[0193] 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 33or can be retrieved using input request 33. Input request 33 can be submitted to model host 31 via an API.

[0194] 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.

[0195] Output payload 34 can include or be based on output(s) 3 from machine- learned model(s) 1. Model host 31 can process output(s) 3 to obtain output payload 34. This can include chaining multiple rounds of inference (e.g., iteratively, recursively, across the same model(s) or different model(s)) to arrive at a final output for a task to be returned in output payload 34. Output payload 34 can be transmitted to client(s) 32 via an API.

[0196] 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.

[0197] Model host 31 can execute machine-learned model(s) 1 to perform inference for various tasks using various types of data. For example, various different input(s) 2 and output(s) 3 can be used for various different tasks. In some implementations, input(s) 2 can be or otherwise represent image data. Machine-learned model(s) 1 can process the image data to generate an output. As an example, machine-learned model(s) 1 can process the image data to generate an image recognition output (e.g., a recognition of the image data, a latent embedding of the image data, an encoded representation of the image data, a hash of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an image segmentation output. As another example, machine-learned model(s) 1 can process the image data to generate an image classification output. As another example, machine-learned model(s) 1 can process the image data to generate an image data modification output (e.g., an alteration of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an encoded image data output (e.g., an encoded and / or compressed representation of the image data, etc.). As another example, machine-learned model(s) 1 can process the image data to generate an upscaled image data output. As another example, machine-learned model(s) 1 can process the image data to generate a prediction output.

[0198] 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.

[0199] In some implementations, input(s) 2 can be or otherwise represent natural language data. Machine-learned model(s) 1 can process the natural language data to generate an output. As an example, machine-learned model(s) 1 can process the natural language data to generate a language encoding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a latent text embedding output. As another example, machine-learned model(s) 1 can process the natural language data to generate a translation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a classification output. As another example, machine-learned model(s) 1 can process the natural language data to generate a textual segmentation output. As another example, machine-learned model(s) 1 can process the natural language data to generate a semantic intent output. As another example, machine-learned model(s) 1 can process the natural language data to generate an upscaled text or natural language output(e.g., text or natural language data that is higher quality than the input text or natural language, etc.). As another example, machine-learned model(s) 1 can process the natural language data to generate a prediction output (e.g., one or more predicted next portions of natural language content).

[0200] 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.

[0201] In some implementations, input(s) 2 can be or otherwise represent latent encoding data (e.g., a latent space representation of an input, etc.). Machine-learned model(s) 1 can process the latent encoding data to generate an output. As an example, machine- learned model(s) 1 can process the latent encoding data to generate a recognition output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a reconstruction output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a search output. As another example, machine- learned model(s) 1 can process the latent encoding data to generate a reclustering output. As another example, machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.

[0202] 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 canprocess 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.

[0203] 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.

[0204] In some implementations, machine-learned model(s) 1 can be configured to perform a task that includes encoding input data for reliable and / or efficient transmission or storage (and / or corresponding decoding). For example, the task may be an audio compression task. The input may include audio data and the output may comprise compressed audio data. In another example, the input includes visual data (e.g. one or more images or videos), the output comprises compressed visual data, and the task is a visual data compression task. In another example, the task may comprise generating an embedding for input data (e.g. input audio or visual data). In some cases, the input includes audio data representing a spoken utterance and the task is a speech recognition task. The output may comprise a text output which is mapped to the spoken utterance. In some cases, the task comprises encrypting or decrypting input data. In some cases, the task comprises a microprocessor performance task, such as branch prediction or memory address translation.

[0205] 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.

[0206] 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.

[0207] 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.

[0208] 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 textualinstructions) 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.

[0209] 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).

[0210] 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).

[0211] 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 togenerate 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

[0212] Figure 11 is a block diagram of an example networked computing system that can perform aspects of example implementations of the present disclosure. The system can include a number of computing devices and systems that are communicatively coupled over a network 49. An example computing device 50 is described to provide an example of a computing device that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). An example server computing system 60 is described as an example of a server computing system that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Computing device 50 and server computing system(s) 60 can cooperatively interact (e.g., over network 49) to perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Model development platform system 70 is an example system that can host or serve model development platform(s) 12 for development of machine-learned models. Third-party system(s) 80 are example system(s) with which any of computing device 50, server computing system(s) 60, or model development platform system(s) 70 can interact in the performance of various aspects of the present disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.).

[0213] Network 49 can be any type of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof and can include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of wired or wireless connection, using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), or protection schemes (e.g., VPN, secure HTTP, SSL). Network 49 can also be implemented via a system bus. For instance, one or more devices or systems of Figure 11 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0214] 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 clientcomputing 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).

[0215] 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.

[0216] 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.

[0217] 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.

[0218] 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.

[0219] 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.

[0220] Server computing system 60 can store or otherwise include one or more machine-learned models 65. Machine-learned model(s) 65 can be the same as or different from machine-learned model(s) 55. Machine-learned models 65 can include one or more machine-learned model(s) 1, such as a sequence processing model 4. Machine-learned models 65 can include one or multiple model instance(s) 31-1. Machine-learned model(s) 65 can be received from computing device 50, model development platform system 70, third party system(s) 80, or developed locally on server computing system(s) 60. Machine-learned model(s) 65 can be loaded into memory 62 and used or otherwise implemented by processor(s) 61. Server computing system(s) 60 can implement multiple parallel instances of machine-learned model (s) 65.

[0221] 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 device50. Machine-learned models 65 can work cooperatively or interoperatively with machine- learned models 55 on computing device 50 to perform various tasks.

[0222] 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.

[0223] Third-party system(s) 80 can include one or more processors 81 and a memory 82. Processor(s) 81 can be any suitable processing device (e.g., a processor core, a microprocessor, an ASIC, an FPGA, a controller, a microcontroller, etc.) and can be one processor or a plurality of processors that are operatively connected. Memory 82 can include one or more non-transitory computer-readable storage media, such as HBM, RAM, ROM, EEPROM, EPROM, flash memory devices, magnetic disks, etc., and combinations thereof. Memory 82 can store data 83 and instructions 84 which can be executed by processor(s) 81 to cause third-party system(s) 80 to perform operations. The operations can implement any one or multiple features described herein. The operations can implement example methods and techniques described herein. Example operations include the functionality described herein with respect to tools and other external resources called when training or performing inference with machine-learned model(s) 1, 4, 16, 20, 55, 65, etc. (e.g., third-party resource(s) 85).

[0224] Figure 11 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update / train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or moretechniques 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).

[0225] Figure 12 is a block diagram of an example computing device 98 that performs according to example embodiments of the present disclosure. Computing device 98 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 98 can include a number of applications (e.g., applications 1 through N). Each application can contain its own machine learning library and machine- learned model(s). For example, each application can include a machine-learned model. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. As illustrated in Figure 12, each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application can communicate with each device component using an API (e.g., a public API). In some implementations, the API used by each application is specific to that application.

[0226] Figure 13 is a block diagram of an example computing device 99 that performs according to example embodiments of the present disclosure. Computing device 99 can be the same as or different from computing device 98. Computing device 99 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 99 can include a number of applications (e.g., applications 1 through N). Each application can be in communication with a central intelligence layer. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. In some implementations, each application can communicate with the central intelligence layer (and model(s) stored therein) using an API (e.g., a common API across all applications).

[0227] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in Figure 13, 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 modelfor 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.

[0228] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in Figure 13, 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

[0229] The technology discussed herein makes reference to servers, databases, software applications, and other computer-based systems, as well as actions taken and information sent to and from such systems. The inherent flexibility of computer-based systems allows for a great variety of possible configurations, combinations, and divisions of tasks and functionality between and among components. For instance, processes discussed herein can be implemented using a single device or component or multiple devices or components working in combination. Databases and applications can be implemented on a single system or distributed across multiple systems. Distributed components can operate sequentially or in parallel.

[0230] While the present subject matter has been described in detail with respect to various specific example embodiments thereof, each example is provided by way of explanation, not limitation of the disclosure. Those skilled in the art, upon attaining an understanding of the foregoing, can readily produce alterations to, variations of, and equivalents to such embodiments. Accordingly, the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that the present disclosure cover such alterations, variations, and equivalents.

[0231] 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 readas limiting the scope of possible combinations of features disclosed herein. Accordingly, the scope of the present disclosure is by way of example rather than by way of limitation, and the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. Moreover, terms are described herein using lists of example elements joined by conjunctions such as “and,” “or,” “but,” etc. It should be understood that such conjunctions are provided for explanatory purposes only. Clauses and other sequences of items joined by a particular conjunction such as “or,” for example, can refer to “and / or,” “at least one of’, “any combination of’ example elements listed therein, etc. Terms such as “based on” should be understood as “based at least in part on.”

[0232] The term “can” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X can perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

[0233] The term “may” should be understood as referring to a possibility of a feature in various implementations and not as prescribing an ability that is necessarily present in every implementation. For example, the phrase “X may perform Y” should be understood as indicating that, in various implementations, X has the potential to be configured to perform Y, and not as indicating that in every instance X must always be able to perform Y. It should be understood that, in various implementations, X might be unable to perform Y and remain within the scope of the present disclosure.

Claims

WHAT IS CLAIMED IS:

1. A computer-implemented method for semantic clustering of sequence data to enable arbitrarily long context windows for machine-learned sequence processing models, comprising: obtaining, by one or more computing devices, a context sequence; determining, by the one or more computing devices, a plurality of subsequences of the context sequence; determining, by the one or more computing devices using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence; determining, by the one or more computing devices, a plurality of semantic clusters, wherein each semantic cluster of the plurality of semantic clusters comprises one or more semantic embeddings of the plurality of semantic embeddings; and generating, by the one or more computing devices using at least one of the first machine-learned sequence processing model and a second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

2. The method of any preceding claim, comprising: selecting, by the one or more computing devices based at least in part on the plurality of semantic clusters, one or more subsequences of the plurality of subsequences of the context sequence; and adding, by the one or more computing devices, data associated with the one or more subsequences to a context window of the second machine-learned sequence processing model; wherein the output sequence is generated based at least in part on the context window.

3. The method of any preceding claim, wherein the data associated with the one or more subsequences comprises at least one of the one or more subsequences.

4. The method of any preceding claim, wherein the data associated with the one or more subsequences comprises sequence data generated by the one or more computing devices based at least in part on a semantic embedding associated with at least one of the one or more subsequences.

5. The method of any preceding claim, wherein the data associated with the one or more subsequences comprises one or more respective semantic embeddings of the plurality of respective semantic embeddings.

6. The method of any preceding claim, wherein the data associated with one or more subsequences comprises sequence data generated by the one or more computing devices based at least in part on a geometric transformation of a semantic embedding associated with at least one of the one or more subsequences.

7. The method of any preceding claim, wherein selecting the one or more subsequences comprises selecting based on one or more priorities.

8. The method of any preceding claim, wherein selecting based on the one or more priorities comprises prioritizing clusters associated with a larger number of semantic embeddings.

9. The method of any preceding claim, wherein selecting based on the one or more priorities comprises prioritizing semantic embeddings that are near a center of a semantic cluster.

10. The method of any preceding claim, wherein the one or more priorities comprise a priority determined based on a user input.

11. The method of any preceding claim, wherein selecting based on the one or more priorities comprises prioritizing a selection of at least one subsequence from each semantic cluster of the plurality of semantic clusters over a selection of a second subsequence from any semantic cluster of the plurality of semantic clusters.

12. The method of any preceding claim, wherein: the first machine-learned model is characterized by a first number of parameters; the second machine-learned model is characterized by a second number of parameters; and the second number is at least 30 times larger than the first number.

13. The method of any preceding claim, wherein determining the plurality of semantic clusters comprises hierarchical clustering.

14. The method of any claim, wherein at least one semantic cluster of the plurality of semantic clusters comprises a plurality of subclusters; and selecting based on the one or more priorities comprises prioritizing semantic embeddings near a centroid of a subcluster of the plurality of subclusters.

15. The method of any preceding claim, wherein determining the plurality of subsequences of the context sequence comprises splitting the context based at least in part on a subsequence length threshold.

16. The method of any preceding claim, wherein determining the plurality of subsequences of the context sequence comprises splitting the context based at least in part on one or more delimiters or sequence boundaries.

17. The method of any preceding claim, wherein determining the plurality of subsequences of the context sequence comprises using a machine-learned model.

18. The method of any preceding claim, further comprising adding a prompt to the context window of the second machine-learned sequence processing model, wherein the prompt comprises one or more instructions or explanations.

19. A computing system comprising one or more processors and one or more non- transitory computer-readable media storing instructions that are executable by the one or more processors to cause the computing system to perform operations, the operations comprising: obtaining a context sequence; determining a plurality of subsequences of the context sequence; determining, using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence; determining a plurality of semantic clusters, wherein each semantic cluster of theplurality of semantic clusters comprises one or more semantic embeddings of the plurality of semantic embeddings; and generating, using at least one of the first machine-learned sequence processing model and a second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

20. One or more non-transitory computer-readable media storing instructions that are executable by a computing system to perform operations, the operations comprising: obtaining a context sequence; determining a plurality of subsequences of the context sequence; determining, using a first machine-learned sequence processing model, a plurality of respective semantic embeddings associated respectively with the plurality of subsequences of the context sequence; determining a plurality of semantic clusters, wherein each semantic cluster of the plurality of semantic clusters comprises one or more semantic embeddings of the plurality of semantic embeddings; and generating, using at least one of the first machine-learned sequence processing model and a second machine-learned sequence processing model, an output sequence based at least in part on the plurality of semantic clusters.

Images