WO2025095958A1 - Downstream adaptations of sequence processing models - Google Patents

Downstream adaptations of sequence processing models Download PDF

Info

Publication number
WO2025095958A1
WO2025095958A1 PCT/US2023/036793 US2023036793W WO2025095958A1 WO 2025095958 A1 WO2025095958 A1 WO 2025095958A1 US 2023036793 W US2023036793 W US 2023036793W WO 2025095958 A1 WO2025095958 A1 WO 2025095958A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
machine
data
training
version
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/US2023/036793
Other languages
French (fr)
Inventor
Victor Carbune
Matthew Sharifi
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Google LLC
Original Assignee
Google LLC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Google LLC filed Critical Google LLC
Priority to PCT/US2023/036793 priority Critical patent/WO2025095958A1/en
Publication of WO2025095958A1 publication Critical patent/WO2025095958A1/en
Anticipated expiration legal-status Critical
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning

Definitions

  • the present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to sequence processing models and adaptations of such models for downstream applications.
  • Artificial intelligence systems increasingly include large foundational machine- learned models which have the capability to provide a wide range of new product experiences. As these large foundational models, also referred to as core models, become more prevalent, so too have adaptations of the systems by downstream users of the models. For instance, downstream users of a pre-trained model may specialize or otherwise modify the model for their own use cases. Different approaches may be used to specialize models for different downstream use cases.
  • one approach may be referred to as parameter- efficient fine-tuning in which a pre-trained model is adapted by changing a small subset of the model weights.
  • Another approach may be referred to as regular fine-tuning in which all or a large number of the model weights are changed to suit a particular use case.
  • Large foundational models are often updated or otherwise modified. For instance, an earlier version of a foundational model may be superseded by a new version of the foundational model. Furthermore, such modification or updating may occur after one or more downstream users have specialized the model (e.g., the earlier version) for their particular use case. Typically, when the foundational model is updated or otherwise modified, the downstream adaptations cannot be preserved.
  • the downstream applications must repeat the original downstream adaptation process on the new version of the core model after it is updated.
  • the downstream adaptation process can involve changing some or all of the model weights of the model (e.g., by further training the model with downstream application specific training data).
  • the computing resource consumption e.g., time and processing capability
  • the computing resource consumption e.g., time and processing capability
  • the computing resource consumption can be significant. This may particularly be the case when there are a large number of downstream applications which each must repeat the downstream adaptation process respectively.
  • One example aspect of the present disclosure is directed to a computer- implemented method that includes, by one or more processors, obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of a machine-learned sequence processing model, generating a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss.
  • Another example aspect of the present disclosure is directed to a system including one or more processors, and one or more computer-readable storage media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations that include obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of a machine-learned sequence processing model, generating a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss.
  • Another example aspect of the present disclosure is directed to one or more non- transitory computer-readable storage media that store at least a portion of a second version of a machine-learned sequence processing model, where the second version of the machine- learned sequence processing is generated by obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of the machine-learned sequence processing model, generating a first training loss based on processing a set of training data by the second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying the at least the portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss.
  • FIG.1 is a block diagram depicting an example computing environment including a host processing system and multiple downstream applications according to example embodiments of the present disclosure
  • FIG.2 is a block diagram depicting an example computing environment including a host sequence processing system and training of a core machine-learned model according to example embodiments of the present disclosure
  • FIG.3 is a block diagram depicting an example computing environment including adapted machine-learned models and a technique for generating a training loss for retraining a core machine-learned model according to example embodiments of the present disclosure
  • FIG.3 is a block diagram depicting an example computing environment including adapted machine-learned models and a technique for generating a training loss for retraining a core machine-learned model according to example embodiments of the present disclosure
  • FIG.12 is a block diagram depicting an example computing environment including adapted machine-learned models and a technique for generating a training loss for retraining a core machine-learned model according to example embodiments of the present disclosure
  • FIG. 4 is a flowchart diagram depicting an example method for training a core machine-learned model using an adapter loss according to example embodiments of the present disclosure
  • FIG. 5 is a flowchart diagram depicting an example method for generating an adapter loss for training a core machine-learned model according to example embodiments of the present disclosure
  • FIG. 6 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure.
  • FIG. 7 is a block diagram of an example processing flow for using machine- learned model(s) to process input(s) to generate output(s) according to example embodiments of the present disclosure
  • FIG.8 is a block diagram of an example sequence processing model according to example embodiments of the present disclosure
  • FIG.9 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example embodiments of the present disclosure
  • FIG.10 is a block diagram of an example model development platform according to example embodiments of the present disclosure
  • FIG.11 is a block diagram of an example training workflow for training a machine-learned model according to example embodiments of the present disclosure
  • FIG.12 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example embodiments of the present disclosure
  • FIG.13 is a block diagram of an example networked computing system according to example embodiments of the present disclosure
  • FIG.13 is a block diagram of an example networked computing system according to example embodiment
  • a foundational or core sequence processing machine-learned (ML) model such as a large language model (LLM) may be adapted by multiple downstream applications to provide specific functionality for each application.
  • ML machine-learned
  • a downstream application can modify or provide additional parameters such as weights and/or layers for the core model that cause the core model to perform differently (e.g., better or in a more preferred manner) by the downstream application for one or more tasks or functions.
  • the core ML model can be retrained or modified without requiring downstream applications to re-run their adaptations of the core ML model.
  • a core machine-learned model can be retrained or modified using one or more loss components that are based on downstream adaptations of the core ML model.
  • the core model can be retrained while preserving the functionality or compatibility of the core model with the downstream applications.
  • the loss component can be generated by processing training data using downstream adaptations of the core processing model. For instance, the loss component can be computed by comparing the output of an adapted version of the model before modification with an output of an adapted version of the model after modification.
  • the loss component can be used to train the new version of the core model to generate an output that is close to the output from the adapted output of the original version of the core model. In this manner, the loss component can be used so that the core model is retrained while preserving the functionality of the downstream adaptation(s) to the core model.
  • the adapter loss component is used in combination with a traditional loss that is determined for the core model during retraining. For example, a set of training data can be provided to the first version of the core model for retraining to generate a second version of the core model. In parallel, the set of training data can be provided to each downstream adapter for the first version of the core model.
  • a core loss component can be determined from the output of the second version of the core model in response to the set of training data.
  • An adapter loss component can be determined based on the outputs of the adapted versions of the core model.
  • the core loss component can be used to update the core model while the adapter loss component can be used to encourage the outputs of the retrained model to remain consistent with the adapted version(s) implemented by downstream applications.
  • the adapter loss component is based on the output of the downstream adapters in combination with the first version of the core model and the output of the downstream adapters in combination with the second version of the core model.
  • the adapter loss component can be used to encourage the output of the downstream adapter with the first version of the model and the output of the downstream adapter with the second version of the model to be the same.
  • a core model such as a core sequence processing model (e.g., large language model)
  • information describing downstream adaptations of the core sequence processing model can be received before modifying a core model such as a core sequence processing model.
  • each of multiple downstream applications may provide a list or other indication of the downstream adaptation of the core model.
  • the downstream adaptation can be implemented by an adapter, which in many instances applies modified or supplemental parameters to the core model.
  • the adapter can specify or include modified weights and/or layers for the core model in some examples. In other examples, the adapter can include additional weights and/or layers for the core model.
  • the downstream applications can provide metadata describing the downstream adapter, such as data identifying a fine-tuning technique that was used to generate the adapter.
  • the core model can be retrained or otherwise updated while using one or more adapter loss components that are based on the downstream adaptation(s) to the core model.
  • the adapter loss component(s) can be generated by processing training data using the downstream adaptations to the core model.
  • the adapter loss component for a particular downstream application can be determined based on the adapted processing by a first version of the core model before retraining and the adapted processing by a second version of the core model that is being retrained.
  • the adapter loss component can force an output of an adaptation of the first version of the core model to be the same as, similar to, or close to an output of an adaptation of the second version of the core model.
  • the core model can be updated while both preserving downstream adaptations and avoiding the introduction of data or other information from the downstream adaptations into the core model. For instance, it is possible that data from a downstream task may leak into a core model through training of the core model using the weights or other information from a downstream application.
  • the second version of the core model can be divided into parameters that are to remain frozen or fixed during retraining, and a subset of parameters that are utilized to make the downstream adapters invariant to retraining of the core model.
  • the training loss can be applied to the subset of parameters that are utilized to make the downstream adapters invariant to retraining. In this way, the training loss can be applied to the subset of parameters used for those adapters so that data does not leak from the adapters back to the core model.
  • the entire second version of the core model can be kept fixed but additional parameters can be created and added so that the invariant parameters of the second version are equal to the summation of the parameters of the second version of the model, the adapted parameters, and the adapter parameters.
  • self-evaluation can be performed to determine whether an adapted output of the first version of the model is strictly worse than the output of the updated core model. If the adapted output is strictly worse, the loss term can be adjusted corresponding to a self-evaluation score between the two outputs. This can lead to either completely ignoring a data point or preserving the adapted behavior if better.
  • the systems and methods can include a computing system that is configured to retrain a machine-learned model while preserving adaptations to the model by downstream applications. By preserving downstream adaptations to the model, the core model can be updated without retraining or otherwise modifying the downstream adaptations to the model.
  • the core model can be updated without additional processing by the downstream applications.
  • the computing resources including processing capacity, power, and storage that can be needed to retrain the downstream adaptations can be saved.
  • the computing resource savings can be significant as the time and processing capability to retrain a downstream adaptation can be significant in many instances.
  • the systems and methods in accordance with the disclosed technology facilitate updating a core model while preserving multiple different adaptations to the core model.
  • different downstream applications may utilize different techniques to generate an adapter for the core model.
  • Some applications may use parameter-efficient fine-tuning such as low-rank adaptation (LoRA) techniques, parameter-efficient fine-tuning (PEFT) techniques, prompt-tuning or custom adapters.
  • LiRA low-rank adaptation
  • PEFT parameter-efficient fine-tuning
  • a core model can be updated using information from downstream adaptations to the core model, while also avoiding data leakage from the downstream applications into the core model.
  • the core model weights can be divided into subsets so that some weights are fixed to avoid data leakage in the upstream direction during retraining.
  • a core model can be updated and leveraged when the updated model performs better on a task than an adapted version.
  • a loss term can be adjusted corresponding to a self-evaluation score between the two outputs so that a data point is either ignored or the adapted behavior is preserved if better.
  • the core sequence processing models can include a large language model (LLM).
  • LLM large language model
  • Much of the following disclosure refers to large language models as specific examples of sequence processing models but it will be appreciated that the disclosure is equally applicable to any type of sequence processing model.
  • the disclosed technology can be used with large image models, multimodal models, and other types of foundational models.
  • the core sequence processing models can operate in domains other than the text domain, such as image domains, audio domains, biochemical domains, etc.
  • a sequence processing model may be used to process sequential inputs for robotic controls and other tasks.
  • the core model and/or the downstream applications can be configured to perform any number of tasks. For instance, if the inputs to the core model and/or a downstream application are images or features that have been extracted from images, the output generated by the core model and/or the downstream application for a given image can be scores for each of a set of object categories, with each score representing an estimated likelihood that the image contains an image of an object belonging to the category.
  • the outputs can be robotic control signals.
  • the output generated can be a score for each of a set of pieces of text, each score representing an estimated likelihood that the piece of text is the correct transcript for the utterance.
  • the output generated may be a score for each of a set of possible diagnoses for the condition of a user, with the score representing an estimated likelihood that the diagnosis is accurate.
  • the output generated may be a score for each of a set of possible responses to the received communication, with the score representing an estimated likelihood that the response matches a user’s intent.
  • the output generated may be a score for each of a set of possible control signals for controlling the apparatus, with the score representing an estimated likelihood that the control signals match the particular function to be performed.
  • the output generated may be a score for each of a set of possible computer readable code segments, with the score representing an estimated likelihood that the computer readable code segments match the computer implemented operation.
  • the output generated may be a score for each of a set of pieces of text in another language, with each score representing an estimated likelihood that the piece of text in the other language is a proper translation of the input text into the other language.
  • FIG. 1 is a block diagram depicting an example computing environment 100 including a host sequence processing system 110 that provides a core sequence processing model that can be accessed by downstream applications.
  • Host sequence processing system implements a first version 115 of a core sequence processing model which is accessed by downstream application 150-1, downstream application 150-2, and downstream application 150-n. Any number of downstream applications may access the first version of the core sequence processing model.
  • host sequence processing system 110 may be implemented by a first computing system and each of the downstream applications 150-1, 150-2, 150-n can be implemented by different remote computing systems.
  • computing environment 100 may be implemented as a client server computing environment, including one or more client computing devices implementing each of the downstream applications and one or more server computing systems implementing host sequence processing system 110.
  • one or more of the downstream applications can be implemented at a server computing system.
  • the computing systems implementing host sequence processing system 110 and downstream applications 150 can be connected by and communicate through one or more networks (not shown). Any number of client computing devices and/or server computing devices can be included in the client-server environment and communicate over a network.
  • the network 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.
  • a client computing device implementing a downstream application can be any suitable device, including, but not limited to, a smartphone, a tablet, a laptop, a desktop computer, or any other computer device that is configured such that it can allow a user to access remote computing devices over a network.
  • the client computing devices can include one or more processor(s), memory, and a display as described in more detail hereinafter.
  • the client computing devices can execute one or more client applications such as a web browser, email application, chat application, video conferencing application, word processing application or the like.
  • the server computing system can include one or more processor(s) and memory implementing host sequence processing system 110.
  • the server computing system can be in communication with the one or more client computing device(s) using a network communication device that is not pictured.
  • system can refer to specialized hardware, computer logic that executes on a more general processor, or some combination thereof.
  • a system can be implemented in hardware, application specific circuits, firmware, and/or software controlling a general-purpose processor.
  • the training or pretraining process can include defining a model architecture (e.g., encoder-decoder, decoder-only, encoder-only, etc.), defining a pre-training procedure (e.g., defining a loss such as masked-language model, causal-language model, mixture of denoisers, etc.), defining a model size (e.g., 1 billion (B) parameters, 8B parameters, 24B parameters, etc.), and defining the pre-training datasets.
  • a model architecture e.g., encoder-decoder, decoder-only, encoder-only, etc.
  • defining a pre-training procedure e.g., defining a loss such as masked-language model, causal-language model, mixture of denoisers, etc.
  • defining a model size e.g., 1 billion (B) parameters, 8B parameters, 24B parameters, etc.
  • a model size e.g., 1 billion (B) parameters, 8B parameters,
  • a first version 115 of the model may be deployed by host sequence processing system 110 (e.g., at a server computing system as part of a cloud computing service) for access and use by downstream applications 150.
  • host sequence processing system 110 e.g., at a server computing system as part of a cloud computing service
  • downstream applications 150 can generate adapters 160 which in combination with the core model, provide adapted models that are specialized for the downstream applications.
  • the first version 115 of the core sequence process model undergoes adaptive training 155-1 by downstream application 150-1 to generate a core sequence model adapter 160-1.
  • the first version 115 of the core sequence process model undergoes adaptive training 155-2 by downstream application 150-2 to generate a core sequence model adapter 160-2, and the first version 115 of the core sequence process model undergoes adaptive training 155-n by downstream application 150-n to generate a core sequence model adapter 160-n.
  • Each adapter 160 can be generated by adaptive training 155.
  • the adaptative training utilized by each downstream application 150-1 can be the same or different.
  • an adapter can be generated by fine-tuning the core model using instruction fine- tuning, human-feedback-fine-tuning via reinforcement learning or direct preference optimization.
  • Other examples of adaptive training can include prompt-tuning of the first version 115 of the core model or neural adapters that can include one more layers.
  • the adapter can perform quantization or other inference runtime optimizations. Embodiments in accordance with the present disclosure can be used with any type of adaptive training and/or adapter.
  • the downstream applications can adapt the weights or other parameters of the core model using downstream datasets in some examples.
  • the downstream datasets can include summarization data, composition data, question/answer data, dialogue data, or any other data that is used to fine-tune the model for a particular downstream application.
  • the downstream application can output entirely new substitute parameter weights for the core model or a subset of substitute parameter weights for the core model.
  • additional or new parameter weights can be added by the adapter.
  • the outputs of the adaptive training 155 by each downstream application results in an adapter 160 that includes the substitute/additional parameter weights, layers, or other data used to adapt the core model to a specific use case for the downstream application.
  • Adaptive training 155-1 by downstream application 150-1 results in a core sequence processing model adapter 160-1
  • adaptive training 155-2 by downstream application 150-2 results in a core sequence processing model adapter 160-2
  • adaptive training 155-n by downstream application 150-n results in a core sequence processing model adapter 160-n.
  • Updating or retraining the first version of the core model as shown at block 120 can include modifying or substituting all or a subset of the weights of the core model, adding additional parameter weights, adding layers to the core model, or any other modification to the parameters defined for the first version of the core model.
  • the retraining results in a second version 125 of the core sequence processing model.
  • the downstream adapters 160 can be dependent upon the weights and other parameters defined for the first version 115 of the core sequence processing model. Typically, modifying the first version 115 of the core sequence processing model will result in incompatibility with the downstream applications.
  • the dependencies of the adapters that refer to modified parameters in the core model can be altered, resulting in the adaptation no longer functioning as desired.
  • FIG. 2 is a block diagram depicting an example computing environment 200 including a host sequence processing system 210 and a technique for updating a core machine-learned model while preserving downstream dependencies or other adaptations according to example embodiments of the present disclosure.
  • FIG. 2 depicts a process for retraining a core sequence processing model to generate a second version 225 of the core sequence processing model after adapters have been created based on a first version of the model.
  • the host sequence processing system 210 can receive information such as data describing downstream adaptations that are to remain compatible with the updates to the core model.
  • a downstream application that is to remain compatible can provide information describing an adapter 260 generated by the downstream application.
  • the data received from each downstream application can include substitute, modified, or additional parameter weights, neural adapter layers, prompts, or other data that defines an adapter for a deployed version of the core sequence processing model.
  • FIG. 2 illustrates three core sequence processing model adapters 260 including core sequence processing model adapters 260-1, 260-2, and 260-n. Data for any number of adapters defined by downstream applications can be received by the host sequence processing system 210. As shown in FIG.
  • the downstream adapters 260 can be instantiated by the host sequence processing system 210 as part of the retraining process.
  • the system utilizes the adapters 260 for the first version of the core model to generate a second version of the core model that remains compatible with the downstream applications.
  • the system can utilize multiple training losses to retrain the first version of the core model and thereby generate a second version of the core model that preserves the downstream adaptations that were generated for the first version of the core model.
  • the system can add one or more loss terms that encourage preservation during training of the desired behavior of the core model as adapted by the downstream applications.
  • Host sequence processing system generates the second version 225 of the core sequence processing model by training that utilizes both a core training loss 230 component and an adapter loss 265 component.
  • the system can provide a set of training data 212 to the second version of the core model and compute a backpropagation based on a core training loss 230 generated by the model in response to the training data.
  • a core loss component can be determined from the output of the second version of the core model in response to the set of training data. Any suitable method can be used to generate the core training loss 230.
  • the core sequence processing model can be a core large language model in example implementations.
  • the set of training data 212 can also be provided to the individual adapters 260 generated by the downstream applications based on the first version of the core model.
  • the system can compute an individual adapter loss 265 separately for each downstream adapter in example implementations. In some examples, a total adapter loss can be determined for all downstream adapters.
  • the individual adapter losses 265 can be combined at 268 (e.g., by summation or other techniques) to generate an aggregate or combined adapter loss.
  • the adapter loss can be determined by processing training data using the downstream adaptations of the core processing model. For instance, the loss component can be computed by comparing the output of an adapted version of the model before modification with an output of an adapted version of the model after modification.
  • the total or combined adapter loss can be used to compute a backpropagation that is used to modify the second version 225 of the core sequence processing model.
  • the individual adapter losses can be used separately to compute a backpropagation for modifying the second version of the core sequence processing model.
  • the loss component can be used to train the new version of the core model to generate an output that is close to the output from the adapted output of the original version of the core model. In this manner, the loss component can be used so that the core model is retrained while preserving the functionality of the downstream adaptation(s) to the core model.
  • the core training loss 230 and the adapter loss(es) 265 can be combined at 269 prior to backpropagation to modify the second version of the core sequence processing model.
  • the loss components can be combined using a summation or other suitable technique.
  • the core training loss 230 and the combined or individual adapter losses 265 can be backpropagated separately.
  • the core loss component can be used to update the core model while the adapter loss component can be used to encourage the outputs of the retrained model to remain consistent with the adapted version(s) implemented by downstream applications.
  • Computing environment 300 depicts one system for processing training data to generate an adapter loss for an adapter.
  • computing environment 300 can be used to generate an adapter loss 265-n for an adapter 260-n as depicted in FIG. 2.
  • Computing environment 300 includes a first adapted model 370 and a second adapted model 372.
  • the first adapted model includes or is otherwise formed by a combination of an adapter 360 and a first version 315 of a core processing model.
  • the adapter 360 can be generated by a downstream application to adapt the first version 315 of the core sequence processing model for a particular application.
  • the second adapted model includes or is otherwise formed by adapter 360 in combination with a second version 325 of the core processing model.
  • adapter 360 can be generated by fine-tuning the first version 315 of the core model.
  • Training data 312 can be provided as an input to the first adapted model 370 and the second adapted model 372.
  • the first adapted model 370 can be formed by the combination of adapter 360 in combination with the first version 315 of the core model.
  • the training data can be provided to the adapted model 370 which includes the sum of the parameters for adapter 360 and the parameters for the first version 315 of the core sequence processing model.
  • the second adapted model 372 can be formed by the combination of adapter 360 and the second version 325 of the core model.
  • the training data can be provided to the adapted model 372 which includes the sum of the parameters for adapter 360 and the parameters for the second version 325 of the core processing model.
  • the first adapted model 370 generates an output 374 and the second adapted model 372 generates an output 376.
  • Output 374 and output 376 are provided to a loss minimization engine 380 which computes an adapter loss 365 for the adapter 360.
  • Adapter loss 365 can then be provided back to the second version of the core sequence processing model by propagation.
  • the system can compute an additional loss for each adapter based on the training data provided to the adapter in combination with the first and second versions of the core model.
  • the system can compute the loss term by minimizing the loss between the output of the first adapted model 370 and the output of the second adapted model 372.
  • the loss term can be a maximum likelihood estimation (MLE) loss in some examples.
  • MLE maximum likelihood estimation
  • KL Kullback–Leibler
  • an MLE loss can include or be equivalent to learning the same output after softmax of logits, while KL divergence can learn to make the logits themselves identical, enforcing the same probability distribution and not just the same prediction.
  • FIG. 4 is a flowchart diagram depicting an example method 400 for training a core machine-learned model using an adapter loss according to example embodiments of the present disclosure.
  • One or more portion(s) of example method 400 and the other methods described herein 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 methods can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example methods 400 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models.
  • the methods in the figures may depict elements performed in a particular order for purposes of illustration and discussion.
  • example method 400 can include obtaining information from one or more downstream applications indicative of adapters implemented by the downstream application(s) for a first version of a core sequence processing model. Various data or information describing downstream adaptations of the core sequence processing model can be received.
  • a downstream application can provide a list, description, or other indication of the downstream adaptation to the core model.
  • the downstream adaptation can include an adapter which is used to apply modified or supplemental parameters to the core model in many instances.
  • the adapter can specify or include modified weights and/or layers for the core model in some examples.
  • the adapter can include additional weights and/or layers for the core model.
  • the downstream applications can provide metadata describing the downstream adapter, such as data identifying a fine-tuning technique that was used to generate the adapter.
  • a downstream application can provide information indicating a subset of parameters from the adapter that are to be maintained during retraining of the core sequence processing model.
  • example method 400 can include obtaining training data for retraining the core sequence processing model.
  • 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.
  • example method 400 can include processing the training data using the core sequence processing model.
  • the training data can be processed using a second version of the core sequence processing model.
  • Individual instances of the training data can be processed to generate an output.
  • the output can be directly obtained from the core machine- learned model or can be a downstream result of a chain of processing operations that includes an output of the core machine-learned model.
  • example method 400 can include processing the training data using each of the adapters for the core sequence processing model. For each adapter, individual instances of the training data can be processed to generate an output.
  • processing the training data by each adapter can include processing the training data by a first adapted model including the adapter in combination with the first version of the core model and processing the training data by a second adapted model including the adapter in combination with the second version of the core model. Each training data instance can be provided to the first adapted model which generates an output and to the second adapted model which generates an output.
  • example method 400 can include determining one or more core training losses based on processing the training data by the second version of the core sequence processing model without any adapters.
  • example method 400 can include determining one or more adapter training losses based on processing the training data by one or more downstream adapters.
  • an adapter training loss can be determined for each downstream adapter and then combined with other adapter training losses.
  • the system can compute a backpropagation based on the individual adapter training losses or a combined adapter training loss.
  • example method 400 can include modifying the second version of the sequence processing model based on the core training loss and the adapter training loss.
  • the core training loss and the adapter training loss can be backpropagated separately to modify the second version of the core sequence model.
  • the core loss and the adapter loss can be combined prior to backpropagation to modify the second version of the core sequence processing model.
  • the individual adapter training losses can be backpropagated separately or combined prior to backpropagation.
  • Backpropagating the loss(es) can include modifying all of the parameters for the second version of the model, modifying a subset of the parameters of the second version of the model, and/or adding new parameters to the second version of the model.
  • FIG. 5 is a flowchart diagram depicting an example method 450 for generating an adapter loss for training a core machine-learned model according to example embodiments of the present disclosure.
  • method 450 is one example of determining an adapter training loss as depicted at 412 in method 400 or an adapter training loss 365 as depicted in FIG. 3.
  • example method 450 can include providing training data as an input to an adapted model that includes an adapter and a first version of a core sequence processing model.
  • the adapter can be generated by a downstream application to adapt the first version of the core sequence processing model for a particular application. For instance, the adapter may be generated by fine-tuning a first version of the core model.
  • the training data can be provided to the adapted model which includes the sum of the parameters for the adapter and the parameters for the first version of the core sequence processing model.
  • example method 450 can include obtaining an output from the first adapted model.
  • example method 450 can include providing the training data as an input to a second adapted model that includes the adapter with a second version of the core sequence processing model.
  • the training data can be provided to the adapted model which includes the sum of the parameters for the adapter and the parameters for the second version of the core sequence processing model.
  • example method 450 can include obtaining an output from the second adapted model that includes the adapter and the second version of the core sequence processing model.
  • the second adapted model generates an output such as one or more inferences.
  • example method 450 can include determining an adapter loss based on the output of the first adapted model and the output of the second adapted model.
  • determining the adapter loss can include minimizing a loss between the output of the first adapted model and the output of the second adapted model to individual instances of the training data.
  • the output of the first adapted model and the output of the second adapted model can be provided to a loss minimization engine which computes an adapter loss for the adapter.
  • the system can compute an additional loss for each adapter based on the training data provided to the adapter in combination with the first and second versions of the core model.
  • the system can compute the loss term by minimizing the loss between the output of the first adapted model and the output of the second adapted model.
  • the loss term can be a maximum likelihood estimation (MLE) loss in some examples.
  • MLE maximum likelihood estimation
  • This additional loss term can be backpropagated to train the second version of the core model while preserving the downstream functionality of the adapter.
  • the additional loss term can force the output of the first adapted model and the second adapted model to be close together. This in turn, encourages the preservation of the behavior of the second version of the core sequence processing model to be equal or similar to the adapted model formed using the first version of the sequence processing model.
  • FIG. 6 depicts a flowchart of a method 500 for training one or more machine- learned models according to aspects of the present disclosure.
  • an example machine-learned model can include a core sequence processing model such as a foundational large language model (LLM).
  • LLM foundational large language model
  • One or more portion(s) of example method 500 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 500 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 500 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models.
  • FIG. 5 depicts elements performed in a particular order for purposes of illustration and discussion.
  • example method 500 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.
  • example method 500 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.
  • example method 500 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function.
  • example method 500 can include updating the machine-learned model using the evaluation signal.
  • 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.
  • 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)).
  • 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.
  • performing backwards propagation of errors can include performing truncated backpropagation through time.
  • Example method 500 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.
  • example method 500 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.).
  • example method 500 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 500 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.
  • example method 500 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.
  • FIG.7 is a block diagram of an example processing flow for using machine- learned model(s) 1 to process input(s) 2 to generate output(s) 3.
  • 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 non-linear 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.
  • 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.
  • machine-learned models can include multi- headed self-attention models.
  • 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.
  • 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).
  • 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.
  • Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g., binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer's central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e.g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like.
  • software code data e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages
  • Data can be raw or processed and can be in any format or schema.
  • 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.
  • 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.
  • FIG.8 is a block diagram of an example implementation of an example machine- learned model configured to process sequences of information.
  • an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4.
  • An example system can pass input(s) 2 to sequence processing model(s) 4.
  • Sequence processing model(s) 4 can include one or more machine-learned components.
  • Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5.
  • Input sequence 5 can include one or more input elements 5-1, 5-2, . . . , 5-M, etc. obtained from input(s) 2.
  • Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7.
  • Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7-N, etc. generated based on input sequence 5.
  • the system can generate output(s) 3 based on output sequence 7.
  • Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models in the text domain are referred to as “Large Language Models,” or LLMs.
  • sequence processing models can operate in other domains, such as image domains, see, e.g., Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXIV:2010.11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et al., MusicLM: Generating Music From Text, ARXIV:2301.11325v1 (Jan.
  • 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. [0104] 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.
  • 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”).
  • 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.
  • elements 5-1, 5-2, . . . , 5-M can represent tokens obtained using a tokenizer.
  • 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.
  • tokenization can be used.
  • textual input source(s) can be tokenized using a byte-pair encoding (BPE) technique.
  • BPE byte-pair encoding
  • SentencePiece A simple and language independent subword tokenizer and detokenizer for Neural Text Processing, PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66–71 (October 31–November 4, 2018), https://aclanthology.org/D18-2012.pdf.
  • Image-based input source(s) can be tokenized by extracting and serializing patches from an image. [0108] In general, arbitrary data types can be serialized and processed into input sequence 5.
  • 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.
  • example prediction layer(s) 6 can predict new output element(s) in view of the context provided by input sequence 5.
  • 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.”
  • a transformer is an example architecture that can be used in prediction layer(s) 6. 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).
  • Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as well as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information.
  • Output sequence 7 can include or otherwise represent the same or different data types as input sequence 5.
  • input sequence 5 can represent textual data
  • output sequence 7 can represent textual data.
  • Input sequence 5 can represent image, audio, or audiovisual data
  • output sequence 7 can represent textual data (e.g., describing the image, audio, or audiovisual data).
  • 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.
  • 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. [0115] 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.
  • output layers e.g., softmax layer
  • 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.
  • 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).
  • Output sequence 7 can include one or multiple portions or elements.
  • 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.).
  • output sequence 7 can include a single element associated with a classification output.
  • an output “vocabulary” can include a set of classes into which an input sequence is to be classified.
  • a vision transformer block can pass latent state information to a multilayer perceptron that outputs a likely class value associated with an input image.
  • FIG.9 is a block diagram of an example technique for populating an example input sequence 8.
  • Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task).
  • Input sequence 8 can include various data elements from different data modalities.
  • 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.
  • Input sequence 8 can be the same as or different from input sequence 5.
  • Input sequence 8 can be a multimodal input sequence that contains elements that represent data from different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions.
  • Input sequence 8 can be configured to contain a plurality of elements that have P dimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween.
  • 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.
  • 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.
  • the expressive power of the embedding space may not be limited to meanings associated with any particular set of tokens or other building blocks.
  • a continuous embedding space can encode a spectrum of high-order information.
  • An individual piece of information e.g., a token
  • An individual piece of information 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.
  • an image patch of an image of a dog on grass can also be projected into the embedding space.
  • 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.
  • the projection of the image patch may not exactly align with any single projection of a single word.
  • 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.
  • 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.
  • 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).
  • the input value represented by element 8-0 can be a learned within a continuous embedding space.
  • 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).
  • 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.
  • a textual data-to-sequence model can subdivide a portion of input text and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-1, 8-2, 8-3, etc.).
  • An image data-to-sequence model can subdivide an input image and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-4, 8-5, 8-6, etc.).
  • An arbitrary datatype data-to-sequence model can subdivide an input of that arbitrary datatype and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7, 8-8, 8-9, etc.).
  • 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.
  • FIG.10 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.).
  • Model 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.
  • 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 pre- trained foundational models 13-1, which can provide a backbone of processing power across various tasks.
  • Model libraries 13 can include one or more pre-trained expert models 13-2, which can be focused on performance in particular domains of expertise.
  • Model libraries 13 can include various model primitives 13-3, which can provide low-level architectures or components (optionally pre-trained), which can be assembled in various arrangements as desired.
  • Model 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.
  • 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.
  • Model alignment toolkit 17 can provide a number of tools for causing development model 16 to generate outputs aligned with desired behavioral characteristics.
  • Alignment can include increasing an accuracy, precision, recall, etc. of model outputs. Alignment can include enforcing output styles, schema, or other preferential characteristics of model outputs. Alignment can be general or domain-specific. For instance, a pre-trained foundational model 13-1 can begin with an initial level of performance across multiple domains. Alignment of the pre-trained foundational model 13-1 can include improving a performance in a particular domain of information or tasks (e.g., even at the expense of performance in another domain of information or tasks). [0131] 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.
  • Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., de- noising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training.
  • Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher- quality data.
  • Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1.
  • Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals.
  • Workbench 15 can implement a fine-tuning pipeline 17-3 to fine- tune development model 16.
  • 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.
  • 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.
  • pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts can include inputs that lack exemplars.
  • 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.
  • 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.
  • 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.
  • 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.
  • deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error.
  • a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool.
  • the tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations.
  • the output of the tool can be returned in response to the original query.
  • 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”).
  • 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.
  • Model plugin toolkit 18 can include interfaces for calling external application programming interfaces (APIs) 18-3.
  • APIs application programming interfaces
  • Model 16 can be aligned to output instruction that initiate API calls to send or obtain data via external systems.
  • 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.
  • 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.
  • 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.
  • hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memory requirements, etc.
  • Tools for distillation 19-3 can provide for the training of lighter-weight models based on the knowledge encoded in development model 16.
  • development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12.
  • 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.
  • 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.
  • FIG.11 is a block diagram of an example training flow for training a machine- learned development model 16.
  • One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices.
  • 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. 10 depicts elements performed in a particular order for purposes of illustration and discussion.
  • 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.
  • Initialized model 21 can undergo pre-training in a pre-training stage 22.
  • Pre- training stage 22 can be implemented using one or more pre-training pipelines 17-2 over data 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).
  • 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.
  • 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.
  • 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.
  • 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.
  • FIG. 12 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.).
  • a model host 31 can receive machine-learned model(s) 1.
  • Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31. [0155] 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.
  • 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.
  • 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.
  • Model host 31 can be implemented by one or multiple computing devices or systems.
  • Client(s) can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31.
  • model host 31 can operate on a server system that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network).
  • Client device(s) can be end-user devices used by individuals.
  • Client device(s) can be server systems that operate client(s) 32 to provide various functionality as a service to downstream end-user devices.
  • 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.
  • model host 31 can be a subroutine or method implemented by one part of an application
  • 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.
  • 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.
  • 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.
  • Input request 33 can include data for input(s) 2.
  • Model host 31 can process input request 33 to obtain input(s) 2.
  • Input(s) 2 can be obtained directly from input request 33 or can be retrieved using input request 33.
  • Input request 33 can be submitted to model host 31 via an API.
  • Model host 31 can perform inference over batches of input requests 33 in parallel.
  • 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.
  • 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.
  • batches of input request(s) 33 can be processed in parallel for higher throughput of output payload(s) 34.
  • 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.
  • Model host 31 can execute machine-learned model(s) 1 to perform inference for various tasks using various types of data.
  • input(s) 2 and output(s) 3 can be used for various different tasks.
  • input(s) 2 can be or otherwise represent image data.
  • Machine-learned model(s) 1 can process the image data to generate an output.
  • 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.).
  • 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.
  • machine-learned model(s) 1 can process the image data to generate an image segmentation output.
  • machine-learned model(s) 1 can process the image data to generate an image classification output.
  • 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.).
  • 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.).
  • machine-learned model(s) 1 can process the image data to generate an upscaled image data output.
  • machine-learned model(s) 1 can process the image data to generate a prediction output.
  • the task is a computer vision task.
  • input(s) 2 includes pixel data for one or more images and the task is an image processing task.
  • 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.
  • 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.
  • the set of categories can be foreground and background.
  • the set of categories can be object classes.
  • 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.
  • 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.
  • 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.
  • machine-learned model(s) 1 can process the natural language data to generate a language encoding output.
  • 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.
  • 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).
  • 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.
  • machine-learned model(s) 1 can process the speech data to generate a speech translation output.
  • machine-learned model(s) 1 can process the speech data to generate a latent embedding output.
  • 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.).
  • 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.).
  • 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.
  • 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.
  • machine-learned model(s) 1 can process the latent encoding data to generate a reconstruction output.
  • machine-learned model(s) 1 can process the latent encoding data to generate a search output.
  • machine-learned model(s) 1 can process the latent encoding data to generate a reclustering output.
  • machine-learned model(s) 1 can process the latent encoding data to generate a prediction output.
  • 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.
  • machine-learned model(s) 1 can process the statistical data to generate a recognition output. As another example, machine-learned model(s) 1 can process the statistical data to generate a prediction output. As another example, machine- learned model(s) 1 can process the statistical data to generate a classification output. As another example, machine-learned model(s) 1 can process the statistical data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the statistical data to generate a visualization output. As another example, machine-learned model(s) 1 can process the statistical data to generate a diagnostic output. [0172] 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.
  • 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.
  • 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).
  • the task may be an audio compression task.
  • the input may include audio data and the output may comprise compressed audio data.
  • 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.
  • the task may comprise generating an embedding for input data (e.g. input audio or visual data).
  • 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.
  • the task comprises encrypting or decrypting input data.
  • the task comprises a microprocessor performance task, such as branch prediction or memory address translation.
  • 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.
  • input(s) 2 can be or otherwise represent data of one or more modalities that encodes context for generating additional content.
  • 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.
  • 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.
  • 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.
  • 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.
  • 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).
  • Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.).
  • One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward answering the question. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of obtaining an answer to the question (e.g., querying a database, performing a computation, executing a script, etc.).
  • 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.
  • 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).
  • 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.
  • 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.
  • FIG. 13 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.).
  • 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.
  • 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.
  • one or more devices or systems of FIG. 12 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.
  • Computing device 50 can be any type of computing device, such as, for example, a personal computing device (e.g., laptop or desktop), a mobile computing device (e.g., smartphone or tablet), a gaming console or controller, a wearable computing device, an embedded computing device, a server computing device, a virtual machine operating on a host device, or any other type of computing device.
  • Computing device 50 can be a client computing device.
  • Computing device 50 can be an end-user computing device.
  • Computing device 50 can be a computing device of a service provided that provides a service to an end user (who may use another computing device to interact with computing device 50).
  • 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.
  • Computing device 50 can also include one or more input components that receive user input.
  • 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.
  • 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.
  • 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.
  • 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.
  • 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.
  • 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.
  • server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50.
  • 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).
  • server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection.
  • computing device 50 can be a workstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e.g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50.
  • 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.
  • 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). [0193] FIG.
  • computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70.
  • computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update/train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or more techniques described herein with respect to model alignment toolkit 17.
  • FIG.14 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.
  • 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).
  • 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.
  • 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.
  • each application can communicate with each device component using an API (e.g., a public API).
  • the API used by each application is specific to that application.
  • FIG.15 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.
  • 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.
  • 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).
  • the central intelligence layer can include a number of machine-learned models. For example, as illustrated in FIG.
  • a respective machine-learned model can be provided for each application and managed by the central intelligence layer.
  • two or more applications can share a single machine-learned model.
  • the central intelligence layer can provide a single model for all of the applications.
  • the central intelligence layer is included within or otherwise implemented by an operating system of computing device 99.
  • the central intelligence layer can communicate with a central device data layer.
  • the central device data layer can be a centralized repository of data for computing device 99. As illustrated in FIG. 15, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components.
  • the central device data layer can communicate with each device component using an API (e.g., a private API).
  • API e.g., a private API.
  • X might be unable to perform Y and remain within the scope of the present disclosure.
  • 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.
  • 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.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Stored Programmes (AREA)

Abstract

Aspects of the disclosed technology include computer-implemented systems and methods for updating machine-learned models while preserving adaptations of the models by downstream applications. The system can obtain data indicative of a respective adapter implemented by each of a plurality of downstream applications for a first version of a machine-learned sequence processing model. The system can generate a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model. The system can generate a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications. The system can modify at least a portion of the second version of the core sequence processing model using the first training loss and the second training loss.

Description

DOWNSTREAM ADAPTATIONS OF SEQUENCE PROCESSING MODELS FIELD [0001] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to sequence processing models and adaptations of such models for downstream applications. BACKGROUND [0002] Artificial intelligence systems increasingly include large foundational machine- learned models which have the capability to provide a wide range of new product experiences. As these large foundational models, also referred to as core models, become more prevalent, so too have adaptations of the systems by downstream users of the models. For instance, downstream users of a pre-trained model may specialize or otherwise modify the model for their own use cases. Different approaches may be used to specialize models for different downstream use cases. For example, one approach may be referred to as parameter- efficient fine-tuning in which a pre-trained model is adapted by changing a small subset of the model weights. Another approach may be referred to as regular fine-tuning in which all or a large number of the model weights are changed to suit a particular use case. [0003] Large foundational models are often updated or otherwise modified. For instance, an earlier version of a foundational model may be superseded by a new version of the foundational model. Furthermore, such modification or updating may occur after one or more downstream users have specialized the model (e.g., the earlier version) for their particular use case. Typically, when the foundational model is updated or otherwise modified, the downstream adaptations cannot be preserved. Instead, the downstream applications must repeat the original downstream adaptation process on the new version of the core model after it is updated. The downstream adaptation process can involve changing some or all of the model weights of the model (e.g., by further training the model with downstream application specific training data). As such, in many instances, the computing resource consumption (e.g., time and processing capability) of repeating the original downstream adaptation process on the new version of the core model after it is updated can be significant. This may particularly be the case when there are a large number of downstream applications which each must repeat the downstream adaptation process respectively. 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] One example aspect of the present disclosure is directed to a computer- implemented method that includes, by one or more processors, obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of a machine-learned sequence processing model, generating a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss. [0006] Another example aspect of the present disclosure is directed to a system including one or more processors, and one or more computer-readable storage media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations that include obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of a machine-learned sequence processing model, generating a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss. [0007] Another example aspect of the present disclosure is directed to one or more non- transitory computer-readable storage media that store at least a portion of a second version of a machine-learned sequence processing model, where the second version of the machine- learned sequence processing is generated by obtaining, from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of the machine-learned sequence processing model, generating a first training loss based on processing a set of training data by the second version of the machine-learned sequence processing model, generating a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, and modifying the at least the portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss. [0008] 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 [0009] FIG.1 is a block diagram depicting an example computing environment including a host processing system and multiple downstream applications according to example embodiments of the present disclosure; [0010] FIG.2 is a block diagram depicting an example computing environment including a host sequence processing system and training of a core machine-learned model according to example embodiments of the present disclosure; [0011] FIG.3 is a block diagram depicting an example computing environment including adapted machine-learned models and a technique for generating a training loss for retraining a core machine-learned model according to example embodiments of the present disclosure; [0012] FIG. 4 is a flowchart diagram depicting an example method for training a core machine-learned model using an adapter loss according to example embodiments of the present disclosure; [0013] FIG. 5 is a flowchart diagram depicting an example method for generating an adapter loss for training a core machine-learned model according to example embodiments of the present disclosure; [0014] FIG. 6 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure; [0015] FIG. 7 is a block diagram of an example processing flow for using machine- learned model(s) to process input(s) to generate output(s) according to example embodiments of the present disclosure; [0016] FIG.8 is a block diagram of an example sequence processing model according to example embodiments of the present disclosure; [0017] FIG.9 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example embodiments of the present disclosure; [0018] FIG.10 is a block diagram of an example model development platform according to example embodiments of the present disclosure; [0019] FIG.11 is a block diagram of an example training workflow for training a machine-learned model according to example embodiments of the present disclosure; [0020] FIG.12 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example embodiments of the present disclosure; [0021] FIG.13 is a block diagram of an example networked computing system according to example embodiments of the present disclosure; [0022] FIG.14 is a block diagram of an example computing device according to example embodiments of the present disclosure; and [0023] FIG.15 is a block diagram of an example computing device according to example embodiments of the present disclosure. DETAILED DESCRIPTION [0024] Reference now will be made in detail to embodiments, one or more examples of which are illustrated in the drawings. Each example is provided by way of explanation of the embodiments, not limitation of the present disclosure. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments without departing from the scope or spirit of the present disclosure. For instance, features illustrated or described as part of one embodiment can be used with another embodiment to yield a still further embodiment. Thus, it is intended that aspects of the present disclosure cover such modifications and variations. Overview [0025] Generally, the present disclosure is directed to machine-learning systems and methods for modifying foundational machine-learned models while preserving downstream adaptations of the foundational core machine-learned models. A foundational or core sequence processing machine-learned (ML) model, such as a large language model (LLM), may be adapted by multiple downstream applications to provide specific functionality for each application. For example, a downstream application can modify or provide additional parameters such as weights and/or layers for the core model that cause the core model to perform differently (e.g., better or in a more preferred manner) by the downstream application for one or more tasks or functions. According to example embodiments of the present disclosure, the core ML model can be retrained or modified without requiring downstream applications to re-run their adaptations of the core ML model. [0026] According to an example implementation of the disclosed technology, a core machine-learned model can be retrained or modified using one or more loss components that are based on downstream adaptations of the core ML model. By using a loss component based on the downstream adaptation(s), the core model can be retrained while preserving the functionality or compatibility of the core model with the downstream applications. [0027] The loss component can be generated by processing training data using downstream adaptations of the core processing model. For instance, the loss component can be computed by comparing the output of an adapted version of the model before modification with an output of an adapted version of the model after modification. The loss component can be used to train the new version of the core model to generate an output that is close to the output from the adapted output of the original version of the core model. In this manner, the loss component can be used so that the core model is retrained while preserving the functionality of the downstream adaptation(s) to the core model. [0028] In some examples, the adapter loss component is used in combination with a traditional loss that is determined for the core model during retraining. For example, a set of training data can be provided to the first version of the core model for retraining to generate a second version of the core model. In parallel, the set of training data can be provided to each downstream adapter for the first version of the core model. A core loss component can be determined from the output of the second version of the core model in response to the set of training data. An adapter loss component can be determined based on the outputs of the adapted versions of the core model. The core loss component can be used to update the core model while the adapter loss component can be used to encourage the outputs of the retrained model to remain consistent with the adapted version(s) implemented by downstream applications. In some examples, the adapter loss component is based on the output of the downstream adapters in combination with the first version of the core model and the output of the downstream adapters in combination with the second version of the core model. The adapter loss component can be used to encourage the output of the downstream adapter with the first version of the model and the output of the downstream adapter with the second version of the model to be the same. [0029] According to an example implementation, before modifying a core model such as a core sequence processing model (e.g., large language model), information describing downstream adaptations of the core sequence processing model can be received. For example, each of multiple downstream applications may provide a list or other indication of the downstream adaptation of the core model. The downstream adaptation can be implemented by an adapter, which in many instances applies modified or supplemental parameters to the core model. The adapter can specify or include modified weights and/or layers for the core model in some examples. In other examples, the adapter can include additional weights and/or layers for the core model. In some examples, the downstream applications can provide metadata describing the downstream adapter, such as data identifying a fine-tuning technique that was used to generate the adapter. [0030] The core model can be retrained or otherwise updated while using one or more adapter loss components that are based on the downstream adaptation(s) to the core model. The adapter loss component(s) can be generated by processing training data using the downstream adaptations to the core model. The adapter loss component for a particular downstream application can be determined based on the adapted processing by a first version of the core model before retraining and the adapted processing by a second version of the core model that is being retrained. The adapter loss component can force an output of an adaptation of the first version of the core model to be the same as, similar to, or close to an output of an adaptation of the second version of the core model. [0031] According to an example implementation, the core model can be updated while both preserving downstream adaptations and avoiding the introduction of data or other information from the downstream adaptations into the core model. For instance, it is possible that data from a downstream task may leak into a core model through training of the core model using the weights or other information from a downstream application. According to example implementations, the second version of the core model can be divided into parameters that are to remain frozen or fixed during retraining, and a subset of parameters that are utilized to make the downstream adapters invariant to retraining of the core model. The training loss can be applied to the subset of parameters that are utilized to make the downstream adapters invariant to retraining. In this way, the training loss can be applied to the subset of parameters used for those adapters so that data does not leak from the adapters back to the core model. In another example, the entire second version of the core model can be kept fixed but additional parameters can be created and added so that the invariant parameters of the second version are equal to the summation of the parameters of the second version of the model, the adapted parameters, and the adapter parameters. [0032] In accordance with aspects of the present disclosure, techniques are provided to leverage instances where a core model after an upgrade performs better on a task than an adapted version of the core model. For example, self-evaluation can be performed to determine whether an adapted output of the first version of the model is strictly worse than the output of the updated core model. If the adapted output is strictly worse, the loss term can be adjusted corresponding to a self-evaluation score between the two outputs. This can lead to either completely ignoring a data point or preserving the adapted behavior if better. [0033] Systems and methods in accordance with example embodiments of the present disclosure provide a number of technical effects and benefits. In particular, the systems and methods can include a computing system that is configured to retrain a machine-learned model while preserving adaptations to the model by downstream applications. By preserving downstream adaptations to the model, the core model can be updated without retraining or otherwise modifying the downstream adaptations to the model. In this manner, the core model can be updated without additional processing by the downstream applications. As such, the computing resources, including processing capacity, power, and storage that can be needed to retrain the downstream adaptations can be saved. The computing resource savings can be significant as the time and processing capability to retrain a downstream adaptation can be significant in many instances. [0034] As an example technical effect and benefit, the systems and methods in accordance with the disclosed technology facilitate updating a core model while preserving multiple different adaptations to the core model. For example, different downstream applications may utilize different techniques to generate an adapter for the core model. Some applications may use parameter-efficient fine-tuning such as low-rank adaptation (LoRA) techniques, parameter-efficient fine-tuning (PEFT) techniques, prompt-tuning or custom adapters. Other applications may use regular fine-tuning by continuing a standard training process using custom data or by leveraging additional signals (e.g., human feedback) for reinforcement learning. In accordance with example embodiments, the parameters for different types of downstream adaptations can be used to preserve the multiple different types of downstream adaptations. In this manner, the systems and methods facilitate different types of downstream adaptations that can each remain compatible with upgrades to the core model. [0035] As another example technical effect and benefit, a core model can be updated using information from downstream adaptations to the core model, while also avoiding data leakage from the downstream applications into the core model. The core model weights can be divided into subsets so that some weights are fixed to avoid data leakage in the upstream direction during retraining. As yet another example technical effect and benefit, a core model can be updated and leveraged when the updated model performs better on a task than an adapted version. A loss term can be adjusted corresponding to a self-evaluation score between the two outputs so that a data point is either ignored or the adapted behavior is preserved if better. [0036] In example implementations, the core sequence processing models can include a large language model (LLM). Much of the following disclosure refers to large language models as specific examples of sequence processing models but it will be appreciated that the disclosure is equally applicable to any type of sequence processing model. For example, the disclosed technology can be used with large image models, multimodal models, and other types of foundational models. For instance, the core sequence processing models can operate in domains other than the text domain, such as image domains, audio domains, biochemical domains, etc. For instance, a sequence processing model may be used to process sequential inputs for robotic controls and other tasks. Similarly, the core model and/or the downstream applications can be configured to perform any number of tasks. For instance, if the inputs to the core model and/or a downstream application are images or features that have been extracted from images, the output generated by the core model and/or the downstream application for a given image can be scores for each of a set of object categories, with each score representing an estimated likelihood that the image contains an image of an object belonging to the category. As another example, if the inputs to the core model and/or a downstream application are sensor data, the outputs can be robotic control signals. [0037] As another example, if the input to the core model and/or a downstream application is a sequence representing a spoken utterance, the output generated can be a score for each of a set of pieces of text, each score representing an estimated likelihood that the piece of text is the correct transcript for the utterance. [0038] As another example, if the input to the core model and/or a downstream application is a sequence of physiological measurements, the output generated may be a score for each of a set of possible diagnoses for the condition of a user, with the score representing an estimated likelihood that the diagnosis is accurate. [0039] As another example, if the input to the core model and/or a downstream application is a sequence of text from a received communication, the output generated may be a score for each of a set of possible responses to the received communication, with the score representing an estimated likelihood that the response matches a user’s intent. [0040] As another example, if the input to the core model and/or a downstream application is indicative of a particular function to be performed by an apparatus (such as a robot), the output generated may be a score for each of a set of possible control signals for controlling the apparatus, with the score representing an estimated likelihood that the control signals match the particular function to be performed. [0041] As another example, if the input to the core model and/or a downstream application includes natural language indicative of a computer implemented operation, the output generated may be a score for each of a set of possible computer readable code segments, with the score representing an estimated likelihood that the computer readable code segments match the computer implemented operation. [0042] As another example, if the input to the core model and/or a downstream application is a sequence of text in one language, the output generated may be a score for each of a set of pieces of text in another language, with each score representing an estimated likelihood that the piece of text in the other language is a proper translation of the input text into the other language. [0043] Although a number of examples of tasks which may be performed by the core model and/or a downstream application are provided here, it will be understood that this is not exhaustive, and that the core model and/or the downstream applications can be configured to perform any suitable task (e.g., including those discussed in relation to FIG. 12). [0044] With reference now to the Figures, example embodiments of the present disclosure will be discussed in further detail. Example Model Arrangements [0045] FIG. 1 is a block diagram depicting an example computing environment 100 including a host sequence processing system 110 that provides a core sequence processing model that can be accessed by downstream applications. Host sequence processing system implements a first version 115 of a core sequence processing model which is accessed by downstream application 150-1, downstream application 150-2, and downstream application 150-n. Any number of downstream applications may access the first version of the core sequence processing model. [0046] In some examples, host sequence processing system 110 may be implemented by a first computing system and each of the downstream applications 150-1, 150-2, 150-n can be implemented by different remote computing systems. For instance, computing environment 100 may be implemented as a client server computing environment, including one or more client computing devices implementing each of the downstream applications and one or more server computing systems implementing host sequence processing system 110. In another example, one or more of the downstream applications can be implemented at a server computing system. [0047] The computing systems implementing host sequence processing system 110 and downstream applications 150 can be connected by and communicate through one or more networks (not shown). Any number of client computing devices and/or server computing devices can be included in the client-server environment and communicate over a network. The network 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. In general, communication between the computing devices can be carried via a network interface using any type of wired and/or wireless connection, using a variety of communication protocols (e.g., TCP/IP, HTTP, RTP, RTCP, etc.), encodings or formats (e.g., HTML, XML, etc.), and/or protection schemes (e.g., VPN, secure HTTP, SSL, etc.). [0048] In some example embodiments, a client computing device implementing a downstream application can be any suitable device, including, but not limited to, a smartphone, a tablet, a laptop, a desktop computer, or any other computer device that is configured such that it can allow a user to access remote computing devices over a network. The client computing devices can include one or more processor(s), memory, and a display as described in more detail hereinafter. The client computing devices can execute one or more client applications such as a web browser, email application, chat application, video conferencing application, word processing application or the like. [0049] The server computing system can include one or more processor(s) and memory implementing host sequence processing system 110. The server computing system can be in communication with the one or more client computing device(s) using a network communication device that is not pictured. [0050] It will be appreciated that the term “system” can refer to specialized hardware, computer logic that executes on a more general processor, or some combination thereof. Thus, a system can be implemented in hardware, application specific circuits, firmware, and/or software controlling a general-purpose processor. In one embodiment, the systems can be implemented as program code files stored on a storage device, loaded into memory and executed by a processor or can be provided from computer program products, for example computer executable instructions, that are stored in a tangible computer-readable storage medium such as RAM, hard disk, or optical or magnetic media. [0051] Host sequence processing system 110 can generate a first version 115 of a core sequence processing model prior to deploying the model for use by downstream applications 150. The first version 115 of the core machine-learned model can be trained and/or pretrained for one or more tasks. Often, a core machine-learned model such as a sequence processing model (e.g., a large language model) is trained and/or pretrained for general functionality so that it can be used for many different downstream applications. [0052] The training or pretraining process can include defining a model architecture (e.g., encoder-decoder, decoder-only, encoder-only, etc.), defining a pre-training procedure (e.g., defining a loss such as masked-language model, causal-language model, mixture of denoisers, etc.), defining a model size (e.g., 1 billion (B) parameters, 8B parameters, 24B parameters, etc.), and defining the pre-training datasets. [0053] As an example, consider a core large language model which may undergo significant training on multiple processors over a long period of time using a large quantity of training data. Such a model may include 1 billion (B), 8B, 16B, 32B, or more parameters. After training, a first version 115 of the model may be deployed by host sequence processing system 110 (e.g., at a server computing system as part of a cloud computing service) for access and use by downstream applications 150. [0054] After initial training and deployment of the first version 115 of the model, downstream applications 150 can generate adapters 160 which in combination with the core model, provide adapted models that are specialized for the downstream applications. In FIG. 1, the first version 115 of the core sequence process model undergoes adaptive training 155-1 by downstream application 150-1 to generate a core sequence model adapter 160-1. The first version 115 of the core sequence process model undergoes adaptive training 155-2 by downstream application 150-2 to generate a core sequence model adapter 160-2, and the first version 115 of the core sequence process model undergoes adaptive training 155-n by downstream application 150-n to generate a core sequence model adapter 160-n. [0055] Each adapter 160 can be generated by adaptive training 155. The adaptative training utilized by each downstream application 150-1 can be the same or different. For example, an adapter can be generated by fine-tuning the core model using instruction fine- tuning, human-feedback-fine-tuning via reinforcement learning or direct preference optimization. Other examples of adaptive training can include prompt-tuning of the first version 115 of the core model or neural adapters that can include one more layers. Additionally, the adapter can perform quantization or other inference runtime optimizations. Embodiments in accordance with the present disclosure can be used with any type of adaptive training and/or adapter. [0056] The downstream applications can adapt the weights or other parameters of the core model using downstream datasets in some examples. For example, the downstream datasets can include summarization data, composition data, question/answer data, dialogue data, or any other data that is used to fine-tune the model for a particular downstream application. As a result of fine-tuning, the downstream application can output entirely new substitute parameter weights for the core model or a subset of substitute parameter weights for the core model. In some examples, additional or new parameter weights can be added by the adapter. The outputs of the adaptive training 155 by each downstream application results in an adapter 160 that includes the substitute/additional parameter weights, layers, or other data used to adapt the core model to a specific use case for the downstream application. Adaptive training 155-1 by downstream application 150-1 results in a core sequence processing model adapter 160-1, adaptive training 155-2 by downstream application 150-2 results in a core sequence processing model adapter 160-2, and adaptive training 155-n by downstream application 150-n results in a core sequence processing model adapter 160-n. [0057] After training and deployment of the first version 115 of the core sequence processing model, and the generation of adapters 160 by downstream applications 150, the host sequence processing system may update the core model by training using additional training data, for example. Updating or retraining the first version of the core model as shown at block 120 can include modifying or substituting all or a subset of the weights of the core model, adding additional parameter weights, adding layers to the core model, or any other modification to the parameters defined for the first version of the core model. The retraining results in a second version 125 of the core sequence processing model. [0058] The downstream adapters 160 can be dependent upon the weights and other parameters defined for the first version 115 of the core sequence processing model. Typically, modifying the first version 115 of the core sequence processing model will result in incompatibility with the downstream applications. The dependencies of the adapters that refer to modified parameters in the core model can be altered, resulting in the adaptation no longer functioning as desired. [0059] In accordance with embodiments of the present disclosure, a second version of the core model can be generated while maintaining compatibility of the core model with the downstream applications, including the adapters for the core model. FIG. 2 is a block diagram depicting an example computing environment 200 including a host sequence processing system 210 and a technique for updating a core machine-learned model while preserving downstream dependencies or other adaptations according to example embodiments of the present disclosure. FIG. 2 depicts a process for retraining a core sequence processing model to generate a second version 225 of the core sequence processing model after adapters have been created based on a first version of the model. [0060] As depicted In FIG. 2, the host sequence processing system 210 can receive information such as data describing downstream adaptations that are to remain compatible with the updates to the core model. A downstream application that is to remain compatible can provide information describing an adapter 260 generated by the downstream application. For example, the data received from each downstream application can include substitute, modified, or additional parameter weights, neural adapter layers, prompts, or other data that defines an adapter for a deployed version of the core sequence processing model. FIG. 2 illustrates three core sequence processing model adapters 260 including core sequence processing model adapters 260-1, 260-2, and 260-n. Data for any number of adapters defined by downstream applications can be received by the host sequence processing system 210. As shown in FIG. 2, the downstream adapters 260 can be instantiated by the host sequence processing system 210 as part of the retraining process. [0061] The system utilizes the adapters 260 for the first version of the core model to generate a second version of the core model that remains compatible with the downstream applications. The system can utilize multiple training losses to retrain the first version of the core model and thereby generate a second version of the core model that preserves the downstream adaptations that were generated for the first version of the core model. In accordance with an example implementation, the system can add one or more loss terms that encourage preservation during training of the desired behavior of the core model as adapted by the downstream applications. [0062] Host sequence processing system generates the second version 225 of the core sequence processing model by training that utilizes both a core training loss 230 component and an adapter loss 265 component. The system can provide a set of training data 212 to the second version of the core model and compute a backpropagation based on a core training loss 230 generated by the model in response to the training data. A core loss component can be determined from the output of the second version of the core model in response to the set of training data. Any suitable method can be used to generate the core training loss 230. The core sequence processing model can be a core large language model in example implementations. [0063] The set of training data 212 can also be provided to the individual adapters 260 generated by the downstream applications based on the first version of the core model. The system can compute an individual adapter loss 265 separately for each downstream adapter in example implementations. In some examples, a total adapter loss can be determined for all downstream adapters. The individual adapter losses 265 can be combined at 268 (e.g., by summation or other techniques) to generate an aggregate or combined adapter loss. [0064] The adapter loss can be determined by processing training data using the downstream adaptations of the core processing model. For instance, the loss component can be computed by comparing the output of an adapted version of the model before modification with an output of an adapted version of the model after modification. The total or combined adapter loss can be used to compute a backpropagation that is used to modify the second version 225 of the core sequence processing model. In some examples, the individual adapter losses can be used separately to compute a backpropagation for modifying the second version of the core sequence processing model. The loss component can be used to train the new version of the core model to generate an output that is close to the output from the adapted output of the original version of the core model. In this manner, the loss component can be used so that the core model is retrained while preserving the functionality of the downstream adaptation(s) to the core model. [0065] As shown in FIG. 2, the core training loss 230 and the adapter loss(es) 265 can be combined at 269 prior to backpropagation to modify the second version of the core sequence processing model. The loss components can be combined using a summation or other suitable technique. In other examples, the core training loss 230 and the combined or individual adapter losses 265 can be backpropagated separately. In various implementations, the core loss component can be used to update the core model while the adapter loss component can be used to encourage the outputs of the retrained model to remain consistent with the adapted version(s) implemented by downstream applications. [0066] FIG. 3 is a block diagram depicting an example computing environment 300 including adapted machine-learned models and a technique for generating a training loss for a core machine-learned model according to example embodiments of the present disclosure. Computing environment depicts one system for processing training data to generate an adapter loss for an adapter. For example, computing environment 300 can be used to generate an adapter loss 265-n for an adapter 260-n as depicted in FIG. 2. [0067] Computing environment 300 includes a first adapted model 370 and a second adapted model 372. The first adapted model includes or is otherwise formed by a combination of an adapter 360 and a first version 315 of a core processing model. The adapter 360 can be generated by a downstream application to adapt the first version 315 of the core sequence processing model for a particular application. The second adapted model includes or is otherwise formed by adapter 360 in combination with a second version 325 of the core processing model. For example, adapter 360 can be generated by fine-tuning the first version 315 of the core model. [0068] Training data 312 can be provided as an input to the first adapted model 370 and the second adapted model 372. The first adapted model 370 can be formed by the combination of adapter 360 in combination with the first version 315 of the core model. Specifically, the training data can be provided to the adapted model 370 which includes the sum of the parameters for adapter 360 and the parameters for the first version 315 of the core sequence processing model. The second adapted model 372 can be formed by the combination of adapter 360 and the second version 325 of the core model. The training data can be provided to the adapted model 372 which includes the sum of the parameters for adapter 360 and the parameters for the second version 325 of the core processing model. [0069] In response to the training data, the first adapted model 370 generates an output 374 and the second adapted model 372 generates an output 376. Output 374 and output 376 are provided to a loss minimization engine 380 which computes an adapter loss 365 for the adapter 360. Adapter loss 365 can then be provided back to the second version of the core sequence processing model by propagation. The system can compute an additional loss for each adapter based on the training data provided to the adapter in combination with the first and second versions of the core model. [0070] As an example, the system can compute the loss term by minimizing the loss between the output of the first adapted model 370 and the output of the second adapted model 372. The loss term can be a maximum likelihood estimation (MLE) loss in some examples. In other examples, a Kullback–Leibler (KL) divergence loss can be used. For instance, an MLE loss can include or be equivalent to learning the same output after softmax of logits, while KL divergence can learn to make the logits themselves identical, enforcing the same probability distribution and not just the same prediction. The additional loss term can be backpropagated to train the second version 325 of the core model while preserving the downstream functionality of the adapter 360. The additional loss term can force the output of the first adapted model using adapter 360 and the output of the second adapted model 372 using adapter 360 to be close together. This in turn, encourages the preservation of the behavior of the second version of the core sequence processing model to be equal or similar to the adapted model 370 formed using the first version of the sequence processing model. Example Methods [0071] FIG. 4 is a flowchart diagram depicting an example method 400 for training a core machine-learned model using an adapter loss according to example embodiments of the present disclosure. One or more portion(s) of example method 400 and the other methods described herein 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 methods can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example methods 400 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. The methods in the figures may depict 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. The example methods are 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 methods can be performed additionally, or alternatively, by other systems. [0072] At 402, example method 400 can include obtaining information from one or more downstream applications indicative of adapters implemented by the downstream application(s) for a first version of a core sequence processing model. Various data or information describing downstream adaptations of the core sequence processing model can be received. A downstream application can provide a list, description, or other indication of the downstream adaptation to the core model. The downstream adaptation can include an adapter which is used to apply modified or supplemental parameters to the core model in many instances. The adapter can specify or include modified weights and/or layers for the core model in some examples. In other examples, the adapter can include additional weights and/or layers for the core model. In some examples, the downstream applications can provide metadata describing the downstream adapter, such as data identifying a fine-tuning technique that was used to generate the adapter. [0073] In some examples, a downstream application can provide information indicating a subset of parameters from the adapter that are to be maintained during retraining of the core sequence processing model. In this manner, the downstream application can limit or otherwise customize which parameters or behaviors of the adapter are to be maintained after training. In some instances, the parameters can be weighted to indicate an importance or flexibility of a particular parameter during retraining. For example, a value of one may indicate that the parameter should be entirely fixed throughout retraining while a value of zero may indicate that the parameter can be freely updated. [0074] At 404, example method 400 can include obtaining training data for retraining the core sequence processing model. 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. [0075] At 406, example method 400 can include processing the training data using the core sequence processing model. The training data can be processed using a second version of the core sequence processing model. Individual instances of the training data can be processed to generate an output. The output can be directly obtained from the core machine- learned model or can be a downstream result of a chain of processing operations that includes an output of the core machine-learned model. [0076] At 408, example method 400 can include processing the training data using each of the adapters for the core sequence processing model. For each adapter, individual instances of the training data can be processed to generate an output. In some examples, processing the training data by each adapter can include processing the training data by a first adapted model including the adapter in combination with the first version of the core model and processing the training data by a second adapted model including the adapter in combination with the second version of the core model. Each training data instance can be provided to the first adapted model which generates an output and to the second adapted model which generates an output. [0077] At 410, example method 400 can include determining one or more core training losses based on processing the training data by the second version of the core sequence processing model without any adapters. In some examples, the system can compute a backpropagation based on a core training loss generated by the model in response to the training data. A core loss component can be determined from the output of the second version of the core model in response to the set of training data without adaptation. Any suitable method can be used to generate the core training loss. [0078] At 412, example method 400 can include determining one or more adapter training losses based on processing the training data by one or more downstream adapters. In some examples, an adapter training loss can be determined for each downstream adapter and then combined with other adapter training losses. The system can compute a backpropagation based on the individual adapter training losses or a combined adapter training loss. [0079] At 414, example method 400 can include modifying the second version of the sequence processing model based on the core training loss and the adapter training loss. In some examples, the core training loss and the adapter training loss can be backpropagated separately to modify the second version of the core sequence model. In some examples, the core loss and the adapter loss can be combined prior to backpropagation to modify the second version of the core sequence processing model. Similarly, the individual adapter training losses can be backpropagated separately or combined prior to backpropagation. Backpropagating the loss(es) can include modifying all of the parameters for the second version of the model, modifying a subset of the parameters of the second version of the model, and/or adding new parameters to the second version of the model. In various implementations, the core loss component can be used to update the core model while the adapter loss component can be used to encourage the outputs of the retrained model to remain consistent with the adapted version(s) implemented by downstream applications. In this manner, the loss component can be used so that the core model is retrained while preserving the functionality of the downstream adaptation(s) to the core model. [0080] FIG. 5 is a flowchart diagram depicting an example method 450 for generating an adapter loss for training a core machine-learned model according to example embodiments of the present disclosure. For example, method 450 is one example of determining an adapter training loss as depicted at 412 in method 400 or an adapter training loss 365 as depicted in FIG. 3. The elements of method 450 may be repeated for each downstream adapter to a core sequence processing model. [0081] At 452, example method 450 can include providing training data as an input to an adapted model that includes an adapter and a first version of a core sequence processing model. The adapter can be generated by a downstream application to adapt the first version of the core sequence processing model for a particular application. For instance, the adapter may be generated by fine-tuning a first version of the core model. The training data can be provided to the adapted model which includes the sum of the parameters for the adapter and the parameters for the first version of the core sequence processing model. [0082] At 454, example method 450 can include obtaining an output from the first adapted model. In response to the training data, the first adapted model generates an output such as one or more inferences. [0083] At 456, example method 450 can include providing the training data as an input to a second adapted model that includes the adapter with a second version of the core sequence processing model. The training data can be provided to the adapted model which includes the sum of the parameters for the adapter and the parameters for the second version of the core sequence processing model. [0084] At 458, example method 450 can include obtaining an output from the second adapted model that includes the adapter and the second version of the core sequence processing model. In response to the training data, the second adapted model generates an output such as one or more inferences. [0085] At 460, example method 450 can include determining an adapter loss based on the output of the first adapted model and the output of the second adapted model. In some examples, determining the adapter loss can include minimizing a loss between the output of the first adapted model and the output of the second adapted model to individual instances of the training data. The output of the first adapted model and the output of the second adapted model can be provided to a loss minimization engine which computes an adapter loss for the adapter. The system can compute an additional loss for each adapter based on the training data provided to the adapter in combination with the first and second versions of the core model. As an example, the system can compute the loss term by minimizing the loss between the output of the first adapted model and the output of the second adapted model. The loss term can be a maximum likelihood estimation (MLE) loss in some examples. This additional loss term can be backpropagated to train the second version of the core model while preserving the downstream functionality of the adapter. The additional loss term can force the output of the first adapted model and the second adapted model to be close together. This in turn, encourages the preservation of the behavior of the second version of the core sequence processing model to be equal or similar to the adapted model formed using the first version of the sequence processing model. [0086] FIG. 6 depicts a flowchart of a method 500 for training one or more machine- learned models according to aspects of the present disclosure. For instance, an example machine-learned model can include a core sequence processing model such as a foundational large language model (LLM). [0087] One or more portion(s) of example method 500 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 500 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 500 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. FIG. 5 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. 5 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 500 can be performed additionally, or alternatively, by other systems. [0088] At 502, example method 500 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 500 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. [0089] At 504, example method 500 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. [0090] At 506, example method 500 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function. Various determinations of loss can be used, such as mean squared error, likelihood loss, cross entropy loss, hinge loss, contrastive loss, or various other loss functions. The evaluation signal can be computed using known ground-truth labels (e.g., supervised learning), predicted or estimated labels (e.g., semi- or self-supervised learning), or without labels (e.g., unsupervised learning). The evaluation signal can be a reward (e.g., for reinforcement learning). The reward can be computed using a machine-learned reward model configured to generate rewards based on output(s) received. The reward can be computed using feedback data describing human feedback on the output(s). [0091] At 508, example method 500 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 500 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained. [0092] In some implementations, example method 500 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.). [0093] In some implementations, example method 500 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 500 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 500 can be implemented for fine-tuning a machine-learned model. Fine-tuning can include, for instance, smaller-scale training on higher-quality (e.g., labeled, curated, etc.) data. Fine-tuning can affect all or a portion of the parameters of a machine-learned model. For example, various portions of the machine-learned model can be “frozen” for certain training stages. For example, parameters associated with an embedding space can be “frozen” during fine-tuning (e.g., to retain information learned from a broader domain(s) than present in the fine-tuning dataset(s)). An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use. Example Machine-Learned Models [0094] FIG.7 is a block diagram of an example processing flow for using machine- learned model(s) 1 to process input(s) 2 to generate output(s) 3. [0095] 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 non-linear 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. [0096] 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 multi- headed self-attention models. [0097] 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). [0098] 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. [0099] Example data types for input(s) 2 or output(s) 3 include natural language text data, software code data (e.g., source code, object code, machine code, or any other form of computer-readable instructions or programming languages), machine code data (e.g., binary code, assembly code, or other forms of machine-readable instructions that can be executed directly by a computer's central processing unit), assembly code data (e.g., low-level programming languages that use symbolic representations of machine code instructions to program a processing unit), genetic data or other chemical or biochemical data, image data, audio data, audiovisual data, haptic data, biometric data, medical data, financial data, statistical data, geographical data, astronomical data, historical data, sensor data generally (e.g., digital or analog values, such as voltage or other absolute or relative level measurement values from a real or artificial input, such as from an audio sensor, light sensor, displacement sensor, etc.), and the like. Data can be raw or processed and can be in any format or schema. [0100] 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. [0101] 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 [0102] FIG.8 is a block diagram of an example implementation of an example machine- learned model configured to process sequences of information. For instance, an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4. An example system can pass input(s) 2 to sequence processing model(s) 4. Sequence processing model(s) 4 can include one or more machine-learned components. Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5. Input sequence 5 can include one or more input elements 5-1, 5-2, . . . , 5-M, etc. obtained from input(s) 2. Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7. Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7-N, etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7. [0103] Sequence processing model(s) 4 can include one or multiple machine-learned model components configured to ingest, generate, or otherwise reason over sequences of information. For example, some example sequence processing models in the text domain are referred to as “Large Language Models,” or LLMs. See, e.g., PaLM 2 Technical Report, GOOGLE, https://ai.google/static/documents/palm2techreport.pdf (n.d.). Other example sequence processing models can operate in other domains, such as image domains, see, e.g., Dosovitskiy et al., An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale, ARXIV:2010.11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et al., MusicLM: Generating Music From Text, ARXIV:2301.11325v1 (Jan. 26, 2023), biochemical domains, see, e.g., Jumper et al., Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example. Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), relatively small models (e.g., fewer parameters, computationally lightweight, etc.), or both. [0104] 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”). [0105] 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. [0106] 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. [0107] For example, elements 5-1, 5-2, . . . , 5-M can represent tokens obtained using a tokenizer. For instance, a tokenizer can process a given portion of an input source and output a series of tokens (e.g., corresponding to input elements 5-1, 5-2, . . . , 5-M) that represent the portion of the input source. Various approaches to tokenization can be used. For instance, textual input source(s) can be tokenized using a byte-pair encoding (BPE) technique. See, e.g., Kudo et al., SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing, PROCEEDINGS OF THE 2018 CONFERENCE ON EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (System Demonstrations), pages 66–71 (October 31–November 4, 2018), https://aclanthology.org/D18-2012.pdf. Image-based input source(s) can be tokenized by extracting and serializing patches from an image. [0108] In general, arbitrary data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-M depicted in FIG. 7 can be the tokens or can be the embedded representations thereof. [0109] 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. [0110] 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.” [0111] A transformer is an example architecture that can be used in prediction layer(s) 6. 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). [0112] Prediction layer(s) 6 can include other machine-learned model architectures in addition to or in lieu of transformer-based architectures. For example, recurrent neural networks (RNNs) and long short-term memory (LSTM) models can also be used, as well as convolutional neural networks (CNNs). In general, prediction layer(s) 6 can leverage various kinds of artificial neural networks that can understand or generate sequences of information. [0113] 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. [0114] 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. [0115] 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. [0116] 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). [0117] Output sequence 7 can include one or multiple portions or elements. In an example content generation configuration, output sequence 7 can include multiple elements corresponding to multiple portions of a generated output sequence (e.g., a textual sentence, values of a discretized waveform, computer code, etc.). In an example classification configuration, output sequence 7 can include a single element associated with a classification output. For instance, an output “vocabulary” can include a set of classes into which an input sequence is to be classified. For instance, a vision transformer block can pass latent state information to a multilayer perceptron that outputs a likely class value associated with an input image. [0118] FIG.9 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements that form part of the model infrastructure, such as an element 8-0 obtained from a task indicator 9 that signals to any model(s) that process input sequence 8 that a particular task is being performed (e.g., to help adapt a performance of the model(s) to that particular task). Input sequence 8 can include various data elements from different data modalities. For instance, an input modality 10-1 can include one modality of data. A data-to-sequence model 11-1 can process data from input modality 10-1 to project the data into a format compatible with input sequence 8 (e.g., one or more vectors dimensioned according to the dimensions of input sequence 8) to obtain elements 8-1, 8-2, 8-3. Another input modality 10-2 can include a different modality of data. A data-to-sequence model 11-2 can project data from input modality 10-2 into a format compatible with input sequence 8 to obtain elements 8-4, 8-5, 8-6. Another input 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. [0119] Input sequence 8 can be the same as or different from input sequence 5. Input sequence 8 can be a multimodal input sequence that contains elements that represent data from different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions. Input sequence 8 can be configured to contain a plurality of elements that have P dimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween. [0120] 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. [0121] 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. [0122] 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. [0123] 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). [0124] Data-to-sequence models 11-1, 11-2, and 11-3 can be the same or different from each other. Data-to-sequence models 11-1, 11-2, and 11-3 can be adapted to each respective input modality 10-1, 10-2, and 10-3. For example, a textual data-to-sequence model can subdivide a portion of input text and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-1, 8-2, 8-3, etc.). An image data-to-sequence model can subdivide an input image and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-4, 8-5, 8-6, etc.). An arbitrary datatype data-to-sequence model can subdivide an input of that arbitrary datatype and project the subdivisions into element(s) in input sequence 8 (e.g., elements 8-7, 8-8, 8-9, etc.). [0125] 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 [0126] FIG.10 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model 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. [0127] 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 pre- trained 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. [0128] 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. [0129] 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. [0130] Model alignment toolkit 17 can provide a number of tools for causing development model 16 to generate outputs aligned with desired behavioral characteristics. Alignment can include increasing an accuracy, precision, recall, etc. of model outputs. Alignment can include enforcing output styles, schema, or other preferential characteristics of model outputs. Alignment can be general or domain-specific. For instance, a pre-trained foundational model 13-1 can begin with an initial level of performance across multiple domains. Alignment of the pre-trained foundational model 13-1 can include improving a performance in a particular domain of information or tasks (e.g., even at the expense of performance in another domain of information or tasks). [0131] 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. [0132] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., de- noising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16. [0133] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher- quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to fine- tune development model 16. [0134] 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. [0135] 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. [0136] In some implementations, pre-trained or fine-tuned models can achieve satisfactory performance without exemplars in the inputs. For instance, zero-shot prompts can include inputs that lack exemplars. Zero-shot prompts can be within a domain within a training dataset or outside of the training domain(s). [0137] 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. [0138] 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. [0139] 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. [0140] 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 500 described above. [0141] Model development platform 12 can include a model plugin toolkit 18. Model plugin toolkit 18 can include a variety of tools configured for augmenting the functionality of a machine-learned model by integrating the machine-learned model with other systems, devices, and software components. For instance, a machine-learned model can use tools to increase performance quality where appropriate. For instance, deterministic tasks can be offloaded to dedicated tools in lieu of probabilistically performing the task with an increased risk of error. For instance, instead of autoregressively predicting the solution to a system of equations, a machine-learned model can recognize a tool to call for obtaining the solution and pass the system of equations to the appropriate tool. The tool can be a traditional system of equations solver that can operate deterministically to resolve the system of equations. The output of the tool can be returned in response to the original query. In this manner, tool use can allow some example models to focus on the strengths of machine-learned models—e.g., understanding an intent in an unstructured request for a task—while augmenting the performance of the model by offloading certain tasks to a more focused tool for rote application of deterministic algorithms to a well-defined problem. [0142] 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”). [0143] 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. [0144] 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. [0145] 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. [0146] Model development platform 12 can include a computational optimization toolkit 19 for optimizing a computational performance of development model 16. For instance, tools for model compression 19-1 can allow development model 16 to be reduced in size while maintaining a desired level of performance. For instance, model compression 19-1 can include quantization workflows, weight pruning and sparsification techniques, etc. Tools for hardware acceleration 19-2 can facilitate the configuration of the model storage and execution formats to operate optimally on different hardware resources. For instance, hardware acceleration 19-2 can include tools for optimally sharding models for distributed processing over multiple processing units for increased bandwidth, lower unified memory requirements, etc. Tools for distillation 19-3 can provide for the training of lighter-weight models based on the knowledge encoded in development model 16. For instance, development model 16 can be a highly performant, large machine-learned model optimized using model development platform 12. To obtain a lightweight model for running in resource-constrained environments, a smaller model can be a “student model” that learns to imitate development model 16 as a “teacher model.” In this manner, for instance, the investment in learning the parameters and configurations of development model 16 can be efficiently transferred to a smaller model for more efficient inference. [0147] 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. [0148] FIG.11 is a block diagram of an example training flow for training a machine- learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other 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. 10 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. 11 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. [0149] 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. [0150] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pre- training stage 22 can be implemented using one or more pre-training pipelines 17-2 over data 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). [0151] 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. [0152] 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. [0153] 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 [0154] FIG. 12 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.). A model host 31 can receive machine-learned model(s) 1. Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31. [0155] 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. [0156] 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. [0157] Model host 31 can be implemented by one or multiple computing devices or systems. Client(s) can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31. [0158] For example, model host 31 can operate on a server system that provides a machine-learning service to client device(s) that operate client(s) 32 (e.g., over a local or wide-area network). Client device(s) can be end-user devices used by individuals. Client device(s) can be server systems that operate client(s) 32 to provide various functionality as a service to downstream end-user devices. [0159] 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. [0160] 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. [0161] 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. [0162] Input request 33 can include data for input(s) 2. Model host 31 can process input request 33 to obtain input(s) 2. Input(s) 2 can be obtained directly from input request 33 or can be retrieved using input request 33. Input request 33 can be submitted to model host 31 via an API. [0163] 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. [0164] 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. [0165] 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. [0166] 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. [0167] 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. [0168] 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). [0169] 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. [0170] 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. [0171] In some implementations, input(s) 2 can be or otherwise represent statistical data. Statistical data can be, represent, or otherwise include data computed and/or calculated from some other data source. Machine-learned model(s) 1 can process the statistical data to generate an output. As an example, machine-learned model(s) 1 can process the statistical data to generate a recognition output. As another example, machine-learned model(s) 1 can process the statistical data to generate a prediction output. As another example, machine- learned model(s) 1 can process the statistical data to generate a classification output. As another example, machine-learned model(s) 1 can process the statistical data to generate a segmentation output. As another example, machine-learned model(s) 1 can process the statistical data to generate a visualization output. As another example, machine-learned model(s) 1 can process the statistical data to generate a diagnostic output. [0172] 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. [0173] 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. [0174] 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. [0175] 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. [0176] 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. [0177] In some implementations, the task can be a question answering task. Machine- learned model(s) 1 can be configured to process input(s) 2 that represent a question to answer and to generate output(s) 3 that advance a goal of returning an answer to the question (e.g., at least a step of a multi-step procedure to perform the function). Output(s) 3 can represent data of the same or of a different modality as input(s) 2. For instance, input(s) 2 can represent textual data (e.g., natural language instructions for a task to be performed) and machine- learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). Input(s) 2 can represent image data (e.g., image-based instructions for a task to be performed, optionally accompanied by textual instructions) and machine-learned model(s) 1 can process input(s) 2 to generate output(s) 3 that represent textual data responsive to the question (e.g., natural language responses, programming language responses, machine language responses, etc.). One or more output(s) 3 can be iteratively or recursively generated to sequentially process and accomplish steps toward answering the question. For instance, an initial output can be executed by an external system or be processed by machine-learned model(s) 1 to complete an initial step of obtaining an answer to the question (e.g., querying a database, performing a computation, executing a script, etc.). Multiple steps can be performed, with a final output being obtained that is responsive to the question. [0178] 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). [0179] 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). [0180] In some implementations, the task can be a data generation task. Machine-learned model(s) 1 can be configured to process input(s) 2 that represent context regarding a desired portion of data (e.g., data from various data domains, such as sensor data, image data, multimodal data, statistical data, etc.). The desired data can be, for instance, synthetic data for training other machine-learned models. The context can include arbitrary data type(s). Machine-learned model(s) 1 can be configured to generate output(s) 3 that represent data that aligns with the desired data. For instance, machine-learned model(s) 1 can be configured to generate data values for populating a dataset. Values for the data object(s) can be selected based on the context (e.g., based on a probability determined based on the context). Example Computing Systems and Devices [0181] FIG. 13 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.). [0182] 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 FIG. 12 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems. [0183] Computing device 50 can be any type of computing device, such as, for example, a personal computing device (e.g., laptop or desktop), a mobile computing device (e.g., smartphone or tablet), a gaming console or controller, a wearable computing device, an embedded computing device, a server computing device, a virtual machine operating on a host device, or any other type of computing device. Computing device 50 can be a client computing device. Computing device 50 can be an end-user computing device. Computing device 50 can be a computing device of a service provided that provides a service to an end user (who may use another computing device to interact with computing device 50). [0184] 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. [0185] 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. [0186] 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. [0187] 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. [0188] 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. [0189] 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. [0190] In an example configuration, machine-learned models 65 can be included in or otherwise stored and implemented by server computing system 60 to establish a client-server relationship with computing device 50 for serving model inferences. For instance, server computing system(s) 60 can implement model host 31 on behalf of client(s) 32 on computing device 50. For instance, machine-learned models 65 can be implemented by server computing system 60 as a portion of a web service (e.g., remote machine-learned model hosting service, such as an online interface for performing machine-learned model operations over a network on server computing system(s) 60). For instance, server computing system(s) 60 can communicate with computing device 50 over a local intranet or internet connection. For instance, computing device 50 can be a workstation or endpoint in communication with server computing system(s) 60, with implementation of machine-learned models 65 being managed by server computing system(s) 60 to remotely perform inference (e.g., for runtime or training operations), with output(s) returned (e.g., cast, streamed, etc.) to computing device 50. Machine-learned models 65 can work cooperatively or interoperatively with machine- learned models 55 on computing device 50 to perform various tasks. [0191] 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. [0192] 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). [0193] FIG. 13 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update/train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or more techniques described herein with respect to model alignment toolkit 17. In this manner, for instance, computing system 50 or server computing system(s) 60 can develop, update/train, or refine machine-learned models based on local datasets (e.g., for model personalization/customization, as permitted by user data preference selections). [0194] FIG.14 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 FIG. 14, each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application 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. [0195] FIG.15 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). [0196] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in FIG. 15, a respective machine-learned model can be provided for each application and managed by the central intelligence layer. In other implementations, two or more applications can share a single machine-learned model. For example, in some implementations, the central intelligence layer can provide a single model for all of the applications. In some implementations, the central intelligence layer is included within or otherwise implemented by an operating system of computing device 99. [0197] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in FIG. 15, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API). Additional Disclosure [0198] 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. [0199] 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. [0200] Aspects of the disclosure have been described in terms of illustrative embodiments thereof. Any and all features in the following claims can be combined or rearranged in any way possible, including combinations of claims not explicitly enumerated in combination together, as the example claim dependencies listed herein should not be read as limiting the scope of possible combinations of features disclosed herein. Accordingly, the scope of the present disclosure is by way of example rather than by way of limitation, and the subject disclosure does not preclude inclusion of such modifications, variations or additions to the present subject matter as would be readily apparent to one of ordinary skill in the art. Moreover, terms are described herein using lists of example elements joined by conjunctions such as “and,” “or,” “but,” etc. It should be understood that such conjunctions are provided for explanatory purposes only. Clauses and other sequences of items joined by a particular conjunction such as “or,” for example, can refer to “and/or,” “at least one of”, “any combination of” example elements listed therein, etc. Terms such as “based on” should be understood as “based at least in part on.” [0201] 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. [0202] 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, comprising: obtaining, by one or more processors from a plurality of downstream applications, data indicative of a respective adapter implemented by each of the plurality of downstream applications for a first version of a machine-learned sequence processing model; generating, by the one or more processors, a first training loss based on processing a set of training data by a second version of the machine-learned sequence processing model; generating, by the one or more processors, a second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications; and modifying, by the one or more processors, at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss.
2. The computer-implemented method of claim 1, wherein generating, by the one or more processors, the second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, comprises: processing the set of training data using a first adapted model comprising the respective adapter implemented by each of the plurality of downstream applications in combination with the first version of the machine-learned sequence processing model; and processing the set of training data using a second adapted model comprising the respective adapter implemented by each of the plurality of downstream applications in combination with the second version of the machine-learned sequence processing model.
3. The computer-implemented method of claim 2, wherein generating, by the one or more processors, the second training loss based on processing the set of training data using the respective adapter implemented by each of the plurality of downstream applications, comprises: minimizing the second training loss based on an output of the first adapted model in response to the set of training data and an output of the second adapted model in response to the set of training data.
4. The computer-implemented method of claim 3, wherein minimizing the second training loss based on the output of the first adapted model in response to the set of training data and the output of the second adapted model in response to the set of training data, comprises: forcing the output of the first adapted model in response to the set of training data and the output of the second adapted model in response to the set of training data to be similar.
5. The computer-implemented method of any of the preceding claims, wherein the second training loss is a maximum likelihood estimation (MLE) loss.
6. The computer-implemented method of any of the preceding claims, wherein the second training loss is a Kullback-Leibler (KL) divergence loss.
7. The computer-implemented method of any of the preceding claims, wherein: the machine-learned sequence processing model is a large foundational model.
8. The computer-implemented method of any of the preceding claims, wherein: the plurality of downstream applications includes a first downstream application that implements a first adapter and a second downstream application that implements a second adapter.
9. The computer-implemented method of claim 8, wherein: the first adapter is generated by fine-tuning the first version of the machine-learned sequence processing model using instruction fine-tuning; and the second adapter is generated by fine-tuning the first version of the machine-learned sequence processing model using human feedback fine-tuning.
10. The computer-implemented method of claim 8, wherein: the first adapter includes at least one of an additional parameter weight for the second version of the machine-learned sequence processing model or a substitute parameter weight for the second version of machine-learned sequence processing model.
11. The computer-implemented method of any of the preceding claims, wherein modifying, by the one or more processors, at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss, comprises: modifying a subset of a plurality of parameters of the second version of the machine- learned sequence processing model; or modifying all of the plurality of parameters of the second version of the machine- learned sequence processing model.
12. The computer-implemented method of any of the preceding claims, wherein modifying, by the one or more processors, at least a portion of the second version of the machine-learned sequence processing model using the first training loss and the second training loss, comprises: adding one or more parameters to the second version of the machine-learned sequence processing model.
13. The computer-implemented method of any of the preceding claims, wherein: the first version and the second version of the machine-learned sequence processing model are implemented by a cloud computing service including a first computing system; and at least one respective adapter is implemented by a second computing system that is remote from and in network communication with the first computing system.
14. A system, comprising: one or more processors; and one or more computer-readable storage media that store instructions that, when executed by the one or more processors, cause the one or more processors to perform the computer-implemented method of any of claims 1-13.
15. One or more non-transitory computer-readable storage media that store the at least the portion of second version of the machine-learned sequence processing model of any of claims 1-13.
PCT/US2023/036793 2023-11-03 2023-11-03 Downstream adaptations of sequence processing models Pending WO2025095958A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/US2023/036793 WO2025095958A1 (en) 2023-11-03 2023-11-03 Downstream adaptations of sequence processing models

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/US2023/036793 WO2025095958A1 (en) 2023-11-03 2023-11-03 Downstream adaptations of sequence processing models

Publications (1)

Publication Number Publication Date
WO2025095958A1 true WO2025095958A1 (en) 2025-05-08

Family

ID=89119502

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2023/036793 Pending WO2025095958A1 (en) 2023-11-03 2023-11-03 Downstream adaptations of sequence processing models

Country Status (1)

Country Link
WO (1) WO2025095958A1 (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20240403566A1 (en) * 2023-06-05 2024-12-05 AIble Inc. Flexible artificial intelligence based system with prompt enhancement
US20250245427A1 (en) * 2024-01-26 2025-07-31 Qualcomm Incorporated Selective parameter-efficient fine-tuning for large-scale models
US20250348667A1 (en) * 2024-05-13 2025-11-13 Nvidia Corporation Generating virtual sensor parameters using large language models for synthetic data generation
US20260050738A1 (en) * 2024-08-14 2026-02-19 Ivalua Sas Use of large language model in source-to-pay platform

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210158837A1 (en) * 2019-11-25 2021-05-27 Qualcomm Incorporated Sound event detection learning

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210158837A1 (en) * 2019-11-25 2021-05-27 Qualcomm Incorporated Sound event detection learning

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
AGOSTINELLI ET AL.: "MusicLM: Genera ing Music From Tex", X :2301.11325V1, 26 January 2023 (2023-01-26)
DOSOVITSKIY ET AL.: "An image is Worth 16x16 Words: Transformers for Image Recognition crt Scale", XIV:2010.1 1929 2, 3 June 2021 (2021-06-03)
JUMPER ET AL.: "Highly accurate protein structure prediction with AlphaFold", NATURE, vol. 596, 26 August 2021 (2021-08-26), pages 583, XP055888904, DOI: 10.1038/s41586-021-03819-2
KUDO: "SentencePiece: A simple and language independent subword tokenizer and cleW kenizen for Neural Text Processing", EMPIRICAL METHODS IN NATURAL LANGUAGE PROCESSING (SYSTEM DEMONSTRATIONS, 4 November 2018 (2018-11-04), pages 66 - 71, Retrieved from the Internet <URL:https://aclanthology.org/Dlβ-2012pdf>
SAHARIA ET AL.: "Non-Autoregressive Machine Translation with Latent Alignments", XIV:2004.07437V3, 16 November 2020 (2020-11-16)
VASWANI ET AL.: "Attention Is All You Need", XIV: 1706.03762V7, 2 August 2023 (2023-08-02)
ZHOU ET AL.: "Mixture-of-Experts with Expert Choice Routing", X)V2202 09368V2, 14 October 2022 (2022-10-14)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20240403566A1 (en) * 2023-06-05 2024-12-05 AIble Inc. Flexible artificial intelligence based system with prompt enhancement
US20250245427A1 (en) * 2024-01-26 2025-07-31 Qualcomm Incorporated Selective parameter-efficient fine-tuning for large-scale models
US20250348667A1 (en) * 2024-05-13 2025-11-13 Nvidia Corporation Generating virtual sensor parameters using large language models for synthetic data generation
US12608543B2 (en) * 2024-05-13 2026-04-21 Nvidia Corporation Generating virtual sensor parameters using large language models for synthetic data generation
US20260050738A1 (en) * 2024-08-14 2026-02-19 Ivalua Sas Use of large language model in source-to-pay platform

Similar Documents

Publication Publication Date Title
WO2024073087A1 (en) Revision of and attribution for output of text generation models
WO2025095958A1 (en) Downstream adaptations of sequence processing models
US20250124256A1 (en) Efficient Knowledge Distillation Framework for Training Machine-Learned Models
WO2024112887A1 (en) Forward-forward training for machine learning
US20250315428A1 (en) Machine-Learning Collaboration System
US20250356223A1 (en) Machine-Learning Systems and Methods for Conversational Recommendations
US20250328568A1 (en) Content-Based Feedback Recommendation Systems and Methods
WO2025101175A1 (en) Llm-centric agile image classification
US20260004490A1 (en) Feedback Predictions for Machine-Learned Generative Models
US20250384346A1 (en) Generation and Processing of Reduced Dimensionality Embeddings
US20250307552A1 (en) Cross-Modal Adapters for Machine-Learned Sequence Processing Models
US20250265087A1 (en) Machine-Learned Model Alignment With Synthetic Data
US20250209355A1 (en) Fast Speculative Decoding Using Multiple Parallel Drafts
WO2025102041A1 (en) User embedding models for personalization of sequence processing models
US20250244960A1 (en) Generative Model Integration with Code Editing
US20250209308A1 (en) Risk Analysis and Visualization for Sequence Processing Models
WO2024207009A1 (en) Efficient use of tools by language models
EP4605858A1 (en) Systems and methods for multi-reward reinforcement learning framework for text-to-image generation
EP4711981A1 (en) Efficient estimation &amp; verification with early exits
US20250131280A1 (en) Meta-Reinforcement Learning Hypertransformers
US20250265285A1 (en) Computing Tool Retrieval Using Sequence Processing Models
US20260119792A1 (en) In-Context Data Synthesis for Machine Learning
US20250124067A1 (en) Method for Text Ranking with Pairwise Ranking Prompting
US12536233B1 (en) AI-generated content page tailored to a specific user
US20260004191A1 (en) Multimodal Machine-Learned Models for Unified Attention and Response Predictions for Visual Content

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23818164

Country of ref document: EP

Kind code of ref document: A1