Tuning text-to-image model utilizing multi-attribute training data and adapted direct performance optimization

Abstract

Systems and methods for tuning text-to-image models utilizing a modified direct performance optimization technique. The method can include generating training data by accessing a number of images associated with content items. The method can include, accessing, for each image, image quality data. The method can include accessing content item performance signal data associated with the content items. The method can include generating preference data from the content item performance signals by selecting a first image and second image based on the image quality data and content item performance signal data. The method can include determining a preferred image based on an analysis of the image quality data and the content item performance signal data. The method can include storing the first and second image as well as the preferred image in a preference data structure and tuning the text-to-image model using the data in the preference data structure.

Description

PCT / US24 / 60996 19 December 2024 (19.12.2024)TUNING TEXT-TO-IMAGE MODEL UTILIZING MULTI- ATTRIBUTE TRAINING DATA AND ADAPTED DIRECT PERFORMANCE OPTIMIZATIONPRIORITY

[0001] The present application claims the benefit of priority of U.S. Provisional Patent Application No. 63 / 698,843, filed on September 25, 2024, which is incorporated by reference herein.FIELD

[0002] The present disclosure relates generally to machine learning processes and machine-learned devices and systems. More particularly, the present disclosure relates to tuning text-to-image models utilizing multi-attribute training data and adapted direct performance optimization techniques.BACKGROUND

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

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

[0005] Example aspects of the present disclosure provide an example method. In some implementations, the example method can include generating, by a computing system, training data by: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content itemPCT / US24 / 60996 19 December 2024 (19.12.2024) performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality data and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning the text-to-image model using the data in the preference data structure.

[0006] Example aspects of the present disclosure provide one or more example non- transitory computer-readable media storing instructions that are executable by one or more processors to cause a computing system to perform example operations. In some implementations, the example operations can include generating, by a computing system, training data by: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content item performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality data and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning the text-to-image model using the data in the preference data structure.

[0007] Example aspects of the present disclosure provide an example computing system that includes one or more processors and one or more example non-transitory computer-readable media storing instructions that are executable by one or more processors to cause a computing system to perform example operations. In some implementations, the example operations can include generating, by a computing system, training data by: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content item performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality dataPCT / US24 / 60996 19 December 2024 (19.12.2024) and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning the text-to-image model using the data in the preference data structure.

[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] Figure 1 depicts a block diagram of an example data flow to perform tuning of a text-to-image generation model according to example embodiments of the present disclosure;

[0010] Figure 2 depicts a block diagram of an example pipeline to perform tuning of a text-to-image generation model according to example embodiments of the present disclosure;

[0011] Figure 3 A depicts an example method for performing tuning of a text-to- image generation model according to example embodiments of the present disclosure;

[0012] Figure 3B depicts an example method for performing tuning of a text-to-image generation model according to example embodiments of the present disclosure;

[0013] Figure 3C depicts an example method for performing tuning of a text-to-image generation model according to example embodiments of the present disclosure;

[0014] Figure 4 is a flow chart diagram illustrating an example method for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0015] Figure 5 is a block diagram of an example processing flow for using machine- learned model(s) to process input(s) to generate output(s) according to example implementations of aspects of the present disclosure;

[0016] Figure 6 is a block diagram of an example sequence processing model according to example implementations of aspects of the present disclosure;PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0017] Figure 7 is a block diagram of an example technique for populating an example input sequence for processing by a sequence processing model according to example implementations of aspects of the present disclosure;

[0018] Figure 8 is a block diagram of an example model development platform according to example implementations of aspects of the present disclosure;

[0019] Figure 9 is a block diagram of an example training workflow for training a machine-learned model according to example implementations of aspects of the present disclosure;

[0020] Figure 10 is a block diagram of an inference system for operating one or more machine-learned model(s) to perform inference according to example implementations of aspects of the present disclosure;

[0021] Figure 11 is a block diagram of an example networked computing system according to example implementations of aspects of the present disclosure;

[0022] Figure 12 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure; and

[0023] Figure 13 is a block diagram of an example computing device according to example implementations of aspects of the present disclosure.DETAILED DESCRIPTION

[0024] Generally, the present disclosure is directed to generating training datasets for use in text-to-image model tuning. In particular, the present disclosure provides for generating training datasets to improve fine-tuning of text-to-image models for multiobjectives such as image quality and performance metrics. This is accomplished using an adapted Direct Preference Optimization (DPO) technique. The adapted approach allows for the model to learn preferences and nuances of content provider-specific quality metrics and performance metrics. As such, the model can be fine-tuned to help optimize performance to align with these metrics.

[0025] The adapted training approach can include tailored prompt expansion. The system can provide for incorporating domain-specific knowledge such as end-user target features or content provider vertical to guide the tailored prompt expansion. As such, the system can provide for expanded prompts that include additional details, context, or instructions to allow for the model to generate images that are likely to have higher provider-PCT / US24 / 60996 19 December 2024 (19.12.2024) specific quality metrics or higher performance metrics than an image that is generated with an initial prompt.

[0026] The adapted training approach can include generating preference data from content item performance signal data and image quality data. The signal data can include multi-attribute continuous or discrete values. As such, a preference for a content provider can be represented as a combination of the respective attributes. The system can provide for a number of approaches for generating preference data for various content providers. The generated preference data can provide for an objective valuation of preferences compared to traditional methods which are subjective. The methods for generating the preference data can include a scalar model or a learning model. The scalar model can provide for weighting all attributes into a single-value preference. As such, a preference can be generated by comparing the scalar preference. The learning model can include training a machine-learned model to merge all attributes into a single embedding vector with a linear layer and logistic activation function. In some instances, the learning model includes a machine-learned model. In some instances, the learning model can include deep learning.

[0027] The adapted training approach can include a step to validate the generated preference data. As will be described herein, the system can utilize DPO to learn to optimize the text-to-image model based on the distribution of the generated preference data. As such, the generated preference data is used for model fine-tuning performance. Thus, the generated preference data can be validated for accuracy or other metrics. In some implementations, the generated preference data can be validated by splitting the content item performance data into a training dataset and a validation set. After a preference is generated, the system can perform an A / B test on the validation data set to confirm that the generated predicted preference data aligns with the actual content item performance data.

[0028] The adapted training approach can include online optimization for the content item pipeline. Existing approaches for utilizing DPO provide for suboptimal performance if the preference data is collected off policy. For instance, existing approaches show that the preference data is associated with images that were not generated by the text-to-image model being tuned with the DPO method provides for suboptimal performance of the text-to-image model in use. As such, the method herein addresses this problem by providing for on-policy data and online learning. On-policy data provides for having preference data that is generated by the same model that is to be optimized. This can provide for benefits including aligning the preference data distribution with native generated data distribution. Online learning can improve the one-shot DPO to continual learning by collecting online content itemPCT / US24 / 60996 19 December 2024 (19.12.2024) performance signal data from the online generation process and continually tune the model based on the data that is collected online. The training data can include both human-labeled training data as well as data that is extracted from content item performance metrics.

[0029] The present disclosure provides for a number of technical effects and benefits. More particularly, the present model provides for a method that improves utilization of finetuning approaches for text-to-image model to improve the quality of the images generated by the text-to-image model by providing for improved training dataset generation as well as improved training and fine-tuning of a text-to-image model. As such, the present disclosure provides for improved training of a text-to-image model to generate images with improved attributes and image quality data. As such, the system can prevent reprocessing of follow-up prompts by generating output images that better align with image quality metrics.

[0030] Additionally, the present disclosure can provide for improved generation of training data for use in tuning the pre-trained text-to-image model for particular downstream metrics including performance signal data and image quality data. Existing methods either consider only one metric or incorrectly weigh the respective metrics in tuning models. The present disclosure addresses these shortcomings in a number of ways, including the manner in which sets or pairs of images are selected to generate preference data. In particular, to generate preference data, the system selects a first and second image based on image quality data, performance signal data, or other data associated with the respective images to select two images that are visually or otherwise close to each other. As such, the preference data generated by comparing the two images to determine a preferred image of the two images can provide a more meaningful update to the text-to-image model by tuning the text-to-image model based on edge cases with more nuanced distinctions opposed to randomly selected images which would not provide as meaningful updates to the text-to-image model. As such, the present disclosure provides a technical solution to this technical problem of generating improved training data for text-to-image generation models such that the text-to-image model can be tuned for a particular use of generating images for use cases associated with a need to improve performance signal data and image quality data associated with the output images.

[0031] Various example implementations are described herein with respect to the accompanying Figures.

[0032] Figure 1 is a block diagram of an example data flow 100. The data flow 100 can be implemented by a computing system that includes one or more computing devices. The data flow 100 can include obtaining input data 102 by a machine-learned model tuning pipeline 114 to output a tuned model 120. Tuned model can process prompt data 122 orPCT / US24 / 60996 19 December 2024 (19.12.2024) expanded prompt data 126 and generate output image(s) 128. Output image(sO 128 can be utilized in content items.

[0033] Input data 102 can include image data 104, original prompt data 106, expanded prompt data 108, performance signal data 110, or image quality data 112. Image data 104 can include one or more images and accompanying data. The accompanying data can include identifiers associated with the image. In some instances, each instance of image data 104 can have associated data from each instance of original prompt data 106, expanded prompt data 108, or performance signal data 110. In some instances, each instance of image data 104 can have associated data from a subset of the original prompt data 106, expanded prompt data 108, or performance signal data 110.

[0034] Original prompt data can include an initial prompt obtained by a machine- learned model, such as a machine-learned model of machine-learned model (s) 118, for generating an output image. Expanded prompt data can include additional details, context, or instructions to allow for the model to generate images that are likely to have higher providerspecific quality metrics or higher performance metrics than an image that is generated with an initial prompt. In some instances, expanded prompt data 108 can be indicative of expanded prompts generated by a prompt expansion pipeline.

[0035] Performance signal data 110 can include data associated with performance of an image or a content item associated with an image. For instance, performance signal data 110 can include data associated with interactions with content items associated with the image or interaction data associated with the image. By way of example, performance signal data 110 can include click through rate, conversion rate, selection rate, or other interaction data associated with an image.

[0036] Image quality data 112 can include a quality score associated with an image. An image quality score can include a discrete or continuous score. The image quality score can be generated by a system based on image attribute data. In some instances, image attribute data can include attribute data generated by a machine-learned model. By way of example, attribute data can include data associated with a resolution of an image, a contrast of the image, a blurriness of the image, a brightness of the image, a saturation of the image, a hue of the image, a sharpness of the image, or a colorfulness of the image. In some instances, attribute data can include a user preference attribute. By way of example, a user preference attribute can include a categorical indication of an image quality. For instance, a categorical indication of an image quality can include at least one of a low quality, a medium quality, or a high quality. In some instances, a categorical indication of an image quality can bePCT / US24 / 60996 19 December 2024 (19.12.2024) generated by a machine-learned model or other algorithm based on the image quality or attribute data.

[0037] Model tuning pipeline 114 can obtain input data 102 to tune model(s) 118. In some implementations, model tuning pipeline 114 can include training data generation pipeline 116. Training data generation pipeline 116 can obtain image data and associated data such as original prompt data 106, expanded prompt data 108, performance signal data 110, or image quality data 112 to generate preference data 124. Preference data 124 can be stored in preference data structure. Preference data 124 can be obtained by model tuning pipeline 114 to tune model(s) 118. Training data generation pipeline 116 will be discussed further with regard to Figure 2.

[0038] Training data generation pipeline 116 can include a preference data generation pipeline to generate preference data 124. The preference data 124 can be generated based on performance signal data 110. Performance signal data 110 can provide useful information as compared to human-labeled preference data. For instance, human-labeled preference data is often binarized whereas performance signal data 110 can include multi -attribute continuous or discrete values. As such, preference data generation pipeline can generate preference data indicative of a combination of a number of attributes.

[0039] The preference value can be generated by a scalar model or a machine-learned model. A preference value can be generated by a scalar model by weighing all attributes into a single-value preference. For instance, a scalar model can generate a single-value preference by performing a mathematical operation to combine or otherwise weigh all the attributes into a single-value preference. In a scalar model implementation, each attribute of a plurality of attributes can be weighted into a single-value preference. The preference can be generated by comparing a number of scalar preferences.

[0040] In a machine-learned model implementation, a preference value can be generated by a machine-learned model based on input data associated with the performance signal data 110.

[0041] DPO can learn to adjust weights or perform other optimizations of machine- learned model(s) 118 based on input data 102 to improve model(s) 118. In some implementations the model(s) can be optimized using off-policy data. Traditional DPO methods provide for suboptimal performance when trained on output data (e.g., image data) that was generated by something other than the model(s) 118 being optimized. The data generated in the preference data generation pipeline can be generated to align with the preference data distribution and the native generated data distribution.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0042] In some implementations, the preference data can be generated by the same model that is going to be optimized. This can provide for improved alignment of the preference data distribution with the generated data distribution.

[0043] In some implementations, the preference data can be generated by an online learning model. The online learning model can collect online content item performance signals from the online generation process and continually improve the model with the data collected. In some instances, the online learning model can include a machine-learned model. In some instances, the online learning model can include a deep learning model.

[0044] Model tuning pipeline 114 can provide for optimizing image generation for content item provider-rated image quality and content item metrics (e.g., customer click- through-rate (CTR), conversion rate, or other advertising metrics). The system can utilize the preference data to fine-tune the text-to-image model using content item performance data. This will enable the content provider team to optimize images generated during the content item performance and content item generation process. As such, an objective can be to optimize image generation for content item provider-rated image quality and content item metrics (e.g., customer click-through-rate (CTR), conversion rate, or other advertising metrics).

[0045] A technical challenge relates to transforming the content item performance dataset into a format suitable for supervised fine-tuning (SFT) and Direct Preference Optimization (DPO). The content item performance dataset can include image data 104, image quality data 112, or performance signal data 110. As described herein image quality data 112 can include an ordinal quality rating such as Low, Good, Best. Performance signal data 110 can include click-through-rate or other metrics. The present disclosure can provide for improvements to existing technology by transforming the content item performance data set into a format suitable for SFT and DPO. Additionally, or alternatively, the present disclosure can provide for checks to prevent output images from containing clickbait or one or more unwanted output artifacts.

[0046] Fine-tuning the model can include generating specific data sets for both supervised fine-tuning and direct preference optimization. By way of example generating specific datasets can include transforming the content item performance data set into a format suitable for SFT and DPO. For instance, supervised fine-tuning datasets can be generated to include and focus on images that have higher performance signal data 110 and higher image quality data 112. Direct performance optimization datasets can be generated based onPCT / US24 / 60996 19 December 2024 (19.12.2024) comparisons of relatively higher and lower quality images within an asset group. As such, the data sets can better be used for their respective intended use cases.

[0047] Dataset generation can include generating text prompts, locating and aggregating competing images, and labeling all examples. Dataset generation will be described in further detail herein.

[0048] In some instances, preference data generation pipeline can include a resolution of a number of data entries associated with a respective image over a number of different contexts of the image. By way of example, an image can be utilized in a number of content item campaigns. Each instance of the image can be associated with distinct performance signal data 110 or image quality data 112. Preference data generation pipeline 118 can reconcile the respective datasets to provide for global optimization and context specific optimization.

[0049] In some implementations, global optimization can be performed. Global optimization can include representing the image preference data score by averaging its performance signal data 110 or image quality data 112 across all instances of the respective image. As such, an image can, in some instances, be determined to be a preferred image and in other contexts an alternative image can be selected as a preferred image.

[0050] In some implementations, context specific optimization can be performed. Context specific optimization can include representing the image preference data score by including context data in the text prompt for the image. By way of example, context data can include content item metadata associated with one or more images. The content item metadata can be utilized to generate a proposed text prompt for recreating the generated image.

[0051] Model tuning pipeline 114 can be utilized to generate a tuned model 120. As described herein, tuned model 120 can include a pre-trained text-to-image model that is finetuned using large volumes of content provider data (e.g., performance signal data 110) and an adapted DPO process. Tuned model 120 can be utilized by one or more sentences to generate output image 128. Output image 128 can be utilized in one or more content item campaigns. In some instances, the system can obtain data including prompt data 122. Prompt data 122 can be natural language input indicative of a request for an image to be generated.

[0052] In some instances, model tuning pipeline 114 can include a combination of supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF), or averaging weights of diffusion models trained on sharded data. SFT can include fine tuning on desired behavior by training a large language model using many prompt-response cases.PCT / US24 / 60996 19 December 2024 (19.12.2024)RLHF can be utilized as described herein to generate DPO training data sets. In some instances, averaging weights of diffusion models trained on sharded data can provide for compartmentalization of various models trained on different shards of data. As such, different models can be tuned based on combining different averages of different diffusion models. This provides for learning and unlearning of data by allowing for different blends of models that have been fine-tuned based on different shards of data.

[0053] Prompt expansion pipeline 124 can obtain prompt data 122 and generate expanded prompt data 126. Expanded prompt data 126 can include context data or instructions to allow for the model to generate images that are likely to have higher providerspecific quality metrics or higher performance metrics than an image that is generated with an initial prompt. In some instances likely can be more than 50%. In some instances, expanded prompt data 126 can be indicative of expanded prompts generated by a prompt expansion pipeline. Expanded prompt data 126 can be utilized to generate output image 128.

[0054] Figure 2 depicts a block diagram of training data generation pipeline 200. Training data generation pipeline 200 can include text prompt generation 205, image sourcing 210, and preference label generation 215.

[0055] Text prompt generation 205 can include generating labels for images using a prompt rewriter or prompt expansion pipeline. In some instances, a machine-learned model can intake image data, process the image, and generate output describing the image. The generated output can be utilized to generate additional images by using the description as a text prompt for a text-to-image model.

[0056] Additionally, or alternatively, the system can heuristically generate a prompt from an existing content item or media asset. By way of example, the system can utilize structured metadata associated with a content item campaign to create a heuristic and convert the metadata into a text prompt. For instance, the metadata can provide additional context for how the image will be used. This can provide for images that are generated to better align with provider-specific quality metrics or performance metrics. In some instances, text prompt generation 205 can include incorporation of domain-specific knowledge into the generation of the prompt. For instance, domain-specific knowledge can include a target audience for a content item, a vertical or market associated with the content item, or other data associated with the context for which the image is being generated. For instance, an image generated for a target audience of children may vary from an image generated for a target audience of adults. Additionally, based on the vertical such a pharmaceutical versus beauty products versus technology versus vehicles, the text prompt and preferences can vary.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0057] Image sourcing 210 can include generating or aggregating competing images. By way of example, images can be generated utilizing the model that is to be tuned or a different text-to-image model. For instance, the system can process the prompt that is generated during text prompt generation 205 and generate an image as output. Additionally, or alternatively, the system can generate pairs of images that are associated with the same content item campaign. As such, the system can generate pairs of images that can be labeled and used to tune the model. For instance, a pair of images can be an initial image generated by the text-to-image model and a second image that has been generated based on the prompt generated by the system. Additionally, or alternatively, the system can generate pairs of images that are in a database and associated with the same content item campaign. These pairs of images and associated data can be labeled and utilized in tuning the text-to-image model.

[0058] Preference label generation 215 can include labeling images or image pairs based on various metrics. The metrics can include, for instance, objective data such as performance signal data or image quality data, human feedback such as a score associated with a content item generated utilizing an image, a weighted score of multiple metrics, a multi-reward reinforcement learning framework, rolling quality-constrained click-through- rate optimization (RQCCO), or any other relevant method for generating or applying preference labels. The labels and data can be utilized to select a preferred image of a pair or set of images. The pair of images, associated data, and the label of the preferred image can be used to tune the text-to-image model to generate images that more closely align with the preferred image.

[0059] In some instances, image quality data can include human feedback. By way of example, the system can obtain data indicative of a score of the image. For instance, the score can be at least one of low, good, or best. In some implementations, the score can be represented numerically (e.g., 1, 2, or 3). The system can utilize the metrics to select a preferred image. The system can select an image with the highest image quality data or score as the preferred image.

[0060] In some instances, performance signal data can include a click-through-rate indicative of whether an image is selected or skipped over. The selection of the image can be indicative of a better quality image. The performance signal data can only be utilized to determine a preferred image if the image has been utilized in a content item campaign and has associated performance signal data. As such, the images cannot be generated on-the-flyPCT / US24 / 60996 19 December 2024 (19.12.2024) but rather must be used in an active content item campaign such that relevant performance signal data can be collected for the image and associated content item.

[0061] In some instances, the system can generate a weighted score of performance signal data and image quality data. This can be performed heuristically by determining an optimal weighting of the metrics in determining an aggregate score. In some instances, the system can process the performance signal data and image quality data by a machine-learned model which can output an aggregated score.

[0062] Preference data can be multi -attribute continuous or discrete values. The preference data generated by the system provides for a preference that utilizes a combination of the attributes. The system can utilize various approaches for generating the preference data. In some implementations, the system can utilize a scalar model or a learning model. The scalar model can provide for weighting all attributes into a single-value preference. The learning model can include training a machine-learned model to merge all attributes into a single embedding vector with a linear layer and logistic activation function. In some instances, the learning model can include a machine-learned model. In some instances, the learning model can include a deep learning model.

[0063] Rolling quality-constrained click-through-rate optimization (RQCCO) can provide for a balance of the selection of a preferred image based on two competing objectives. For instance, the two competing objectives can include a primary object that needs maximization, such as a click-through -rate, and a secondary objective with a minimum threshold, such as an image quality score. The training data generation pipeline 200 can maintain a rolling score for a weighted average quality of the chosen preferred images (e.g., a rolling quality score. If incorporating a new image keeps the rolling quality score above a predefined threshold, T, the model can prioritize the primary objective (e.g., performance signal data) even if the new item has a lower quality score (e.g., image quality score).

[0064] By way of example, RQCCO can utilize the following definitions:Let Qt represent the image quality score of image i.Let CTRi represent the click-through rate of image i.Let yi be the model’s selection indicator for image i (1 if selected, 0 otherwise) Let Q represent the rolling quality score.Let T represent a quality threshold.

[0065] The objective can be set at maximizing the click-through rate (CTR) subject to maintaining a quality score above the threshold, T. The RQCCO rule can be calculating the new rolling quality score if an image is selected using the following formula:PCT / US24 / 60996 19 December 2024 (19.12.2024) n - Q + QjQ new n + 1

[0066] Where n represents the number of images selected before the current calculation is performed.

[0067] The selection condition can be set that If Qnewis greater than or equal to the threshold, then the image with the highest CTR is selected.

[0068] By way of example, a machine-learned model can evaluate two images in a DPO fine tuning process wherein the first image has a CTR = 9.37, Quality Score = Low (1.0) and a second image can have a CTR = 2.23, Quality Score = Best (3.0). The current rolling average Quality Score can be 2.7. The Quality Score Threshold, T, can be 2.5. The previous number of example images, n, can be 999. Using the formula above, the first image has a higher CTR and results in decreasing the current rolling average Quality score from 2.7 to 2.6983. This new current rolling average quality score is still above the threshold, T. Because this condition is met and the CTR of the first image is greater than that of the second image, the image will be selected as the preferred image of the set of the first image and the second image.

[0069] In some implementations, the training data generation pipeline 200 can include preference validation. By way of example, DPO can learn to optimize the text-to- image model to align with the preference data distribution. As such, generation of the preference data distribution is important for the model fine-tuning performance. This can be performed by generating a training dataset and a validation dataset from the performance signal data. After a preference is generated, the system can facilitate A / B test on the validation set to confirm that the generated preference aligns with the performance signal data (e.g., positively impacts metrics associated with a content item campaign).

[0070] In some implementations, on policy data can be obtained and the text-to-image model can be optimized on-line using at least one of on-policy data or online learning. On- policy data can include preference data generated by the same model (e.g., text-to-image mode) that is being optimized. This can provide for aligning image quality data distribution alongside native generated data distributions. Online learning can improve over existing DPO methods by providing for continual learning. As such, the system can collect performance signal data associated with content item campaigns from the online generation pipeline and continue to tune and improve the model based on the collected data.

[0071] Figure 3A, Figure 3B, and Figure 3C depict flow diagrams of example method 300 to perform tuning text-to-image model utilizing multi-attribute training data and adaptedPCT / US24 / 60996 19 December 2024 (19.12.2024) direct performance optimization according to example embodiments of the present disclosure. The method 300 can be performed by processing logic that can include hardware (e.g., processing device, circuitry, dedicated logic, programmable logic, microcode, hardware of a device, integrated circuit, etc.), software (e.g., instructions run or executed on a processing device), or a combination thereof. In some embodiments, method 300 is performed by a server computing system (e.g., server computing system 60) or client computing system (e.g., client computing device 50). Although shown in a particular sequence or order, unless otherwise specified, the order of the processes can be modified. Thus, the illustrated embodiments should be understood only as examples, and the illustrated processes can be performed in a different order, and some processes can be performed in parallel. Additionally, one or more processors can be omitted in various embodiments. Thus, not all processes are required in every embodiment. Other process flows are possible.

[0072] At operation 302, processing logic can generate training data. For instance, processing logic can generate the training data by performing operations described in Figure 3B.

[0073] At operation 302 A, processing logic can access a plurality of images associated with a plurality of content items. Each image of the plurality of images can have associated data. In some instances, the data can include an indication of one or more content item campaigns associated with the image or content item. Data can additionally, or alternatively, include prompt data, such as an initial text prompt or expanded prompt data such as a prompt that was expanded upon by a system to generate the image.

[0074] At operation 302B, processing logic can access, for each image of the plurality of images, image quality data. As described herein, image quality data can be determined based on attribute data associated with the image. The plurality of attributes associated with the image can include at least one of: a resolution of the image, a contrast of the image, a blurriness of the image, a brightness of the image, a saturation of the image, a hue of the image, a sharpness of the image, or a colorfulness of the image. Additionally, or alternatively, the attribute data can include a user preference attribute. By way of example, a user preference attribute can include a categorical indication of an image quality. For instance, a categorical indication of an image quality can include at least one of a low quality, a medium quality, or a high quality. In some instances, a categorical indication of an image quality can be generated by a machine-learned model or other algorithm based on the image quality or attribute data.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0075] In some instances, image quality data can include a quality score generated by a machine-learned model configured to generate a quality score for an image based on a plurality of attributes associated with the image. In some implementations, the plurality of attributes associated with the image can be generated by a machine-learned model configured to generate a quality score for an image based on the plurality of attributes associated with the image. The image quality score can include a discrete or continuous score. A discrete score can include a categorical indication of an image quality. A continuous score can include a numerical score, such as a numerical value within a range.

[0076] In some instances, the image quality data can include a plurality of instances of the first image. In some instances, processing logic can generate an average preference value for the first image based on a plurality of instances of the first image. In some instances, processing logic can generate a plurality of instances of the first image. The preference data for each respective instance of the plurality of instances of the first image can include context data.

[0077] For instance, an image can be associated with a number of content items or discrete content item campaigns. As such, there can be a number of data entries associated with the image with differing image data, prompt data, image quality data, or performance signal data. As such, the system must determine which of the entries to utilize and whether the entries should be combined and averaged. As described herein, there can be both global optimization as well as context-specific optimization. The methods described herein provide for determining whether a global or context-specific optimization should be used and generating the training data accordingly (e.g., determining whether the data set will be used for SFT, DPO, or both).

[0078] At operation 302C, processing logic can access a plurality of content item performance signal data associated with the plurality of content items. Content item performance signal data can include at least one of a click-through-rate associated with the content item or a conversion rate associated with the content item. A click-through-rate can be indicative of whether a user selected the image associated with the content item. A conversion rate can be indicative of whether a user performed an action associated with the content item. The action can include, for instance, visiting a secondary website, downloading an application, subscribing to a newsletter, creating an account, purchasing an item associated with the content item, or some other target action.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0079] At operation 304, processing logic can generate preference data from the content item performance signals. For instance, processing logic can generate preference data by performing operations described in Figure 3C.

[0080] At operation 304 A, processing logic can select a first image and a second image based on the image quality data and content item performance signal data. For instance, processing logic can select images that are similar or have been utilized in similar content item campaigns. As such, pairs of images can be selected such that selection between the two images can provide for valuable learning for tuning the text-to-image model.Compared to traditional methods which may select random images to compare, this approach enables the provision of more valuable input data to tune the model such that the ultimate output of the model is better tailored to the image quality and performance signal metrics associated with the images generated by the text-to-image model.

[0081] At operation 304B, processing logic can determine a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data. As described herein, the preferred image can be selected based on comparison of multiple attributes associated with the respective images. In some instances, determining the preferred image can be based on performance signal data or image quality data, human feedback such as a score associated with a content item generated utilizing an image, a weighted score of multiple metrics, a multi-reward reinforcement learning framework, rolling quality-constrained click-through-rate optimization (RQCCO), or any other relevant method for generating or applying preference labels.

[0082] In some implementations, to determine a preferred image of the first image and the second image, processing logic can generate a first preference value for the first image and a second preference value for the second image based on the image quality data and the content item performance signal data.

[0083] In some implementations, to determine a preferred image of the first image and the second image, processing logic can determine a first image as the preferred image based on a comparison of the first preference value and the second preference value.

[0084] In some embodiments, the first preference value and the second preference value can be generated by a scalar model. The scalar model can be configured to generate a preference value for an image based on a weighted combination of the image quality data and the content item performance signal data. By selecting a preferred image based on their respective image quality and content item performance signal data, the present solution provides valuable information that can help mapping performance with quality metrics acrossPCT / US24 / 60996 19 December 2024 (19.12.2024) pairs of images. Such information can be advantageously used to train a text-to-image model to improve the quality of the generated images.

[0085] In some embodiments, the first preference value and the second preference value are generated by a machine-learned model. By way of example, the machine-learned model can generate the first preference value and the second preference value based on combining the image quality data and the content item data for the respective images into a single embedding vector for each respective image.

[0086] At operation 304C, processing logic can store the selected first image and second image and the preferred image in a preference data structure. For instance, the preference data structure can be any structure for storing data associated with image pairs, associated image data, and a preferred image of the pair of images. In some instances, the training data can be aggregated and a distribution for the training data can be stored and utilized for tuning the text-to-image model.

[0087] At operation 306, processing logic can tune the text-to-image model using the data in the preference data structure. Tuning the text-to-image model can include a combination of supervised fine-tuning (SFT) and direct performance optimization (DPO). SFT can be used to improve the model's general image generation quality, while DPO can be used to specifically optimize the model for the desired content item metrics. This combined approach can lead to a more effective and efficient tuning process.

[0088] The tuned text-to-image model can be utilized to generate images or content items to be utilized in content item campaigns. After a content item has been utilized for a period of time in a content item campaign, the system can collect performance signal data to use to continually tune the text-to-image model as described herein.

[0089] Figure 4 depicts a flowchart of a method 400 for training one or more machine-learned models according to aspects of the present disclosure. For instance, an example machine-learned model can include a text-to-image generation model or a prompt expansion model.

[0090] One or more portion(s) of example method 400 can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of example method 400 can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of example method 400 can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. Figure 4 depicts elements performed in a particularPCT / US24 / 60996 19 December 2024 (19.12.2024) order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. Figure 4 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of example method 400 can be performed additionally, or alternatively, by other systems.

[0091] At 402, example method 400 can include obtaining a training instance. A set of training data can include a plurality of training instances divided between multiple datasets (e.g., a training dataset, a validation dataset, or testing dataset). A training instance can be labeled or unlabeled. Although referred to in example method 400 as a “training” instance, it is to be understood that runtime inferences can form training instances when a model is trained using an evaluation of the model’s performance on that runtime instance (e.g., online training / learning). Example data types for the training instance and various tasks associated therewith are described throughout the present disclosure.

[0092] At 404, example method 400 can include processing, using one or more machine-learned models, the training instance to generate an output. The output can be directly obtained from the one or more machine-learned models or can be a downstream result of a chain of processing operations that includes an output of the one or more machine- learned models.

[0093] At 406, example method 400 can include receiving an evaluation signal associated with the output. The evaluation signal can be obtained using a loss function. Various determinations of loss can be used, such as mean squared error, likelihood loss, cross entropy loss, hinge loss, contrastive loss, or various other loss functions. The evaluation signal can be computed using known ground-truth labels (e.g., supervised learning), 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).

[0094] At 408, example method 400 can include updating the machine-learned model using the evaluation signal. For example, values for parameters of the machine-learned model(s) can be learned, in some embodiments, using various training or learning techniques, such as, for example, backwards propagation. For example, the evaluation signal can bePCT / US24 / 60996 19 December 2024 (19.12.2024) backpropagated from the output (or another source of the evaluation signal) through the machine-learned model(s) to update one or more parameters of the model(s) (e.g., based on a gradient of the evaluation signal with respect to the parameter value(s)). For example, system(s) containing one or more machine-learned models can be trained in an end-to-end manner. Gradient descent techniques can be used to iteratively update the parameters over a number of training iterations. In some implementations, performing backwards propagation of errors can include performing truncated backpropagation through time. Example method 400 can include implementing a number of generalization techniques (e.g., weight decays, dropouts, etc.) to improve the generalization capability of the models being trained.

[0095] In some implementations, example method 400 can be implemented for training a machine-learned model from an initialized state to a fully trained state (e.g., when the model exhibits a desired performance profile, such as based on accuracy, precision, recall, etc.).

[0096] In some implementations, example method 400 can be implemented for particular stages of a training procedure. For instance, in some implementations, example method 400 can be implemented for pre-training a machine-learned model. Pre-training can include, for instance, large-scale training over potentially noisy data to achieve a broad base of performance levels across a variety of tasks / data types.

[0097] In some implementations, example method 400 can be implemented for finetuning a machine-learned model. Fine-tuning can include, for instance, smaller-scale training on higher-quality (e.g., labeled, curated, etc.) data. Fine-tuning can affect all or a portion of the parameters of a machine-learned model. For example, various portions of the machine- learned model can be “frozen” for certain training stages. For example, parameters associated with an embedding space can be “frozen” during fine-tuning (e.g., to retain information learned from a broader domain(s) than present in the fine-tuning dataset(s)). In some implementations, example method 400 uses adapter modules. Adapters can be small trainable layers that are inserted between pre-existing layers of a pre-trained model. During the finetuning process, the original parameters of the pre-trained model are typically frozen, and only the parameters of the adapters are updated.

[0098] In some implementations, example method 400 can be implemented to execute parameter-efficient fine-tuning methods, such as Layerwise Optimization of Residuals (LoRA). LoRA can refine pre-trained models with minimal adjustments to the original parameters. This can be achieved by introducing trainable low-rank matrices that modify the behavior of the pre-trained weights without directly altering them. In somePCT / US24 / 60996 19 December 2024 (19.12.2024) implementations, during fine-tuning, only these auxiliary matrices are updated, which significantly reduces the number of parameters that are trained.

[0099] An example fine-tuning approach includes reinforcement learning. Reinforcement learning can be based on user feedback on model performance during use.

[0100] Figure 5 is a block diagram of an example processing flow for using machine- learned model(s) 1 to process input(s) 2 to generate output(s) 3.

[0101] Machine-learned model(s) 1 can be or include one or multiple machine- learned models or model components. Example machine-learned models can include neural networks (e.g., deep neural networks). Example machine-learned models can include nonlinear models or linear models. Example machine-learned models can use other architectures in lieu of or in addition to neural networks. Example machine-learned models can include decision tree based models, support vector machines, hidden Markov models, Bayesian networks, linear regression models, k-means clustering models, etc.

[0102] Machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of the machine-learned models described above with respect to the preceding figures. For example, machine-learned model(s) 1 can be or include, or otherwise be representative of any one or more of include a text-to-image generation model or a prompt expansion model, etc. Although various features, variations, and implementations described below are described with respect to machine-learned model(s) 1, it is to be understood that such features, variations, and implementations are to be understood as described with respect to each of include a text-to-image generation model or a prompt expansion model, etc., any other machine-learned component described herein.

[0103] Example neural networks can include feed-forward neural networks, recurrent neural networks (RNNs), including long short-term memory (LSTM) based recurrent neural networks, convolutional neural networks (CNNs), diffusion models, generative-adversarial networks, or other forms of neural networks. Example neural networks can be deep neural networks. Some example machine-learned models can leverage an attention mechanism such as self-attention. For example, some example machine-learned models can include multiheaded self-attention models.

[0104] Machine-learned model(s) 1 can include a single or multiple instances of the same model configured to operate on data from input(s) 2. Machine-learned model(s) 1 can include multiple different models or multiple different model portions configured to operate on data from input(s) 2.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0105] Machine-learned model(s) 1 can include an ensemble of different models that can cooperatively interact to process data from input(s) 2. For example, a model ensemble can include multiple models that have different attributes (e.g., different architectures, trained with different recipes, etc.). The ensemble can output an overall output based on the individual outputs of the constituent models. In this manner, for instance, the diverse constituent models can work together to provide system-level robustness by effectively aggregating over individual strengths and weaknesses of any given model. The respective individual outputs can be combined in a weighted combination, using a voting or routing mechanism, or a learned output layer (e.g., one or more feedforward or fully-connected layers).

[0106] 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). For example, different portions of a model can learn (explicitly or implicitly) different expertise areas, with pathways through the model being selected by a learned routing mechanism that engages the appropriate expert for a given input (e.g., a given portion of an input, such as on a per-token basis). For example, a feedforward network can be sparsely activated for a given portion of an input based on an output of a routing mechanism that processes the portion of the input. In this manner, for instance, the group of activated weights can form an “expert” that is selected by the router. On each forward pass, only a subset of the total model weights may be engaged, thereby decreasing a quantity of operations performed for processing a given input compared to a densely activated model. In this manner, for instance, the expressive and interpretive power of a high-parameter-count model can be achieved with more compute-efficient forward passes.

[0107] 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.

[0108] 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,PCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0109] 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.

[0110] 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.

[0111] Figure 6 is a block diagram of an example implementation of an example machine-learned model configured to process sequences of information. For instance, an example implementation of machine-learned model(s) 1 can include machine-learned sequence processing model(s) 4. An example system can pass input(s) 2 to sequence processing model(s) 4. Sequence processing model(s) 4 can include one or more machine- learned components. Sequence processing model(s) 4 can process the data from input(s) 2 to obtain an input sequence 5. Input sequence 5 can include one or more input elements 5-1, 5- 2, . . . , 5-A , etc. obtained from input(s) 2. Sequence processing model 4 can process input sequence 5 using prediction layer(s) 6 to generate an output sequence 7. Output sequence 7 can include one or more output elements 7-1, 7-2, . . . , 7 -A, etc. generated based on input sequence 5. The system can generate output(s) 3 based on output sequence 7.

[0112] 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 atPCT / US24 / 60996 19 December 2024 (19.12.2024)Scale, ARXIV:2010.11929v2 (Jun. 3, 2021), audio domains, see, e.g., Agostinelli et al., MusicLM: Generating Music From Text, ARXIV:2301.1 1325V1 (Jan. 26, 2023), biochemical domains, see, e.g., Jumper et al., Highly accurate protein structure prediction with AlphaFold, 596 Nature 583 (Aug. 26, 2021), by way of example. Sequence processing model(s) 4 can process one or multiple types of data simultaneously. Sequence processing model(s) 4 can include relatively large models (e.g., more parameters, computationally expensive, etc.), relatively small models (e.g., fewer parameters, computationally lightweight, etc.), or both.

[0113] 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”).

[0114] 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.

[0115] 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.

[0116] For example, elements 5-1, 5-2, . . . , 5M 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, . . . , 5M) 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0117] In general, arbitrary data types can be serialized and processed into input sequence 5. It is to be understood that element(s) 5-1, 5-2, . . . , 5-M depicted in Figure 6 can be the tokens or can be the embedded representations thereof.

[0118] 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.

[0119] 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.”

[0120] 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).

[0121] 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.

[0122] 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, orPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0123] 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.

[0124] 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.

[0125] 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).

[0126] 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.

[0127] Figure 7 is a block diagram of an example technique for populating an example input sequence 8. Input sequence 8 can include various functional elements thatPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0128] Input sequence 8 can be the same as or different from input sequence 5. Input sequence 8 can be a multimodal input sequence that contains elements that represent data from different modalities using a common dimensional representation. For instance, an embedding space can have P dimensions. Input sequence 8 can be configured to contain a plurality of elements that have / Jdimensions. In this manner, for instance, example implementations can facilitate information extraction and reasoning across diverse data modalities by projecting data into elements in the same embedding space for comparison, combination, or other computations therebetween.

[0129] 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.

[0130] 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 valuePCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0131] 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.

[0132] 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).

[0133] 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.).

[0134] 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 processingPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0135] Figure 8 is a block diagram of an example model development platform 12 that can facilitate creation, adaptation, and refinement of example machine-learned models (e.g., machine-learned model(s) 1, sequence processing model(s) 4, etc.). Model development platform 12 can provide a number of different toolkits that developer systems can employ in the development of new or adapted machine-learned models.

[0136] Model development platform 12 can provide one or more model libraries 13 containing building blocks for new models. Model libraries 13 can include one or more pretrained foundational models 13-1, which can provide a backbone of processing power across various tasks. Model libraries 13 can include one or more pre-trained expert models 13-2, which can be focused on performance in particular domains of expertise. Model libraries 13 can include various model primitives 13-3, which can provide low-level architectures or components (optionally pre-trained), which can be assembled in various arrangements as desired. Model primitives 13-3 can include a library of pre-trained adapters or LoRA modules that can adapt a baseline foundational model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like.

[0137] 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.

[0138] 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.

[0139] 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).PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0140] 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.

[0141] Pre-training pipelines 17-2 can include a machine-learned model training workflow configured to update development model 16 over large-scale, potentially noisy datasets. For example, pre-training can leverage unsupervised learning techniques (e.g., denoising, etc.) to process large numbers of training instances to update model parameters from an initialized state and achieve a desired baseline performance. Pre-training pipelines 17-2 can leverage unlabeled datasets in dataset(s) 17-1 to perform pre-training. Workbench 15 can implement a pre-training pipeline 17-2 to pre-train development model 16.

[0142] Fine-tuning pipelines 17-3 can include a machine-learned model training workflow configured to refine the model parameters of development model 16 with higher- quality data. Fine-tuning pipelines 17-3 can update development model 16 by conducting supervised training with labeled dataset(s) in dataset(s) 17-1. Fine-tuning pipelines 17-3 can update development model 16 by conducting reinforcement learning using reward signals from user feedback signals. Workbench 15 can implement a fine-tuning pipeline 17-3 to finetune development model 16.

[0143] 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.

[0144] 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.

[0145] 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).

[0146] 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., inputPCT / US24 / 60996 19 December 2024 (19.12.2024) element values) based on one or more training iterations. Workbench 15 can implement prompt engineering tools in development model 16.

[0147] Prompt libraries 17-4 can include pipelines for prompt generation. For example, inputs can be generated using development model 16 itself or other machine- learned models. In this manner, for instance, a first model can process information about a task and output an input for a second model to process in order to perform a step of the task. The second model can be the same as or different from the first model. Workbench 15 can implement prompt generation pipelines in development model 16.

[0148] 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.

[0149] Although various training examples described herein with respect to model development platform 12 refer to “pre-training” and “fine-tuning,” it is to be understood that model alignment toolkit 17 can generally support a wide variety of training techniques adapted for training a wide variety of machine-learned models. Example training techniques can correspond to the example training method 400 described above.

[0150] 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0151] 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”).

[0152] 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.

[0153] Model plugin toolkit 18 can include interfaces for calling external application programming interfaces (APIs) 18-3. For instance, in addition to or in lieu of implementing tool calls or tool code directly with development model 16, development model 16 can be aligned to output instructions that initiate API calls to send or obtain data via external systems.

[0154] 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.

[0155] 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 toPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0156] 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.

[0157] Figure 9 is a block diagram of an example training flow for training a machine-learned development model 16. One or more portion(s) of the example training flow can be implemented by a computing system that includes one or more computing devices such as, for example, computing systems described with reference to the other figures. Each respective portion of the example training flow can be performed by any (or any combination) of one or more computing devices. Moreover, one or more portion(s) of the example training flow can be implemented on the hardware components of the device(s) described herein, for example, to train one or more systems or models. FIG. 9 depicts elements performed in a particular order for purposes of illustration and discussion. Those of ordinary skill in the art, using the disclosures provided herein, will understand that the elements of any of the methods discussed herein can be adapted, rearranged, expanded, omitted, combined, or modified in various ways without deviating from the scope of the present disclosure. FIG. 9 is described with reference to elements / terms described with respect to other systems and figures for exemplary illustrated purposes and is not meant to be limiting. One or more portions of the example training flow can be performed additionally, or alternatively, by other systems.

[0158] 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.

[0159] Initialized model 21 can undergo pre-training in a pre-training stage 22. Pretraining stage 22 can be implemented using one or more pre-training pipelines 17-2 over data from dataset(s) 17-1. Pre-training can be omitted, for example, if initialized model 21 is already pre-trained (e.g., development model 16 contains, is, or is based on a pre-trained foundational model or an expert model).PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0160] Pre-trained model 23 can then be a new version of development model 16, which can persist as development model 16 or as a new development model. Pre-trained model 23 can be the initial state if development model 16 was already pre-trained. Pre-trained model 23 can undergo fine-tuning in a fine-tuning stage 24. Fine-tuning stage 24 can be implemented using one or more fine-tuning pipelines 17-3 over data from dataset(s) 17-1. Fine-tuning can be omitted, for example, if a pre-trained model has satisfactory performance, if the model was already fine-tuned, or if other tuning approaches are preferred.

[0161] 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 of 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.

[0162] 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.

[0163] Figure 10 is a block diagram of an inference system for operating one or more machine-learned model(s) 1 to perform inference (e.g., for training, for deployment, etc.). A model host 31 can receive machine-learned model(s) 1. Model host 31 can host one or more model instance(s) 31-1, which can be one or multiple instances of one or multiple models. Model host 31 can host model instance(s) 31-1 using available compute resources 31-2 associated with model host 31.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0164] 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.

[0165] 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.

[0166] Model host 31 can be implemented by one or multiple computing devices or systems. Client(s) 2 can be implemented by one or multiple computing devices or systems, which can include computing devices or systems shared with model host 31.

[0167] 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.

[0168] 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 orPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0169] Model instance(s) 31-1 can include one or more machine-learned models that are available for performing inference. Model instance(s) 31-1 can include weights or other model components that are stored in persistent storage, temporarily cached, or loaded into high-speed memory. Model instance(s) 31-1 can include multiple instance(s) of the same model (e.g., for parallel execution of more requests on the same model). Model instance(s) 31-1 can include instance(s) of different model(s). Model instance(s) 31-1 can include cached intermediate states of active or inactive model(s) used to accelerate inference of those models. For instance, an inference session with a particular model may generate significant amounts of computational results that can be re-used for future inference runs (e.g., using a KV cache for transformer-based models). These computational results can be saved in association with that inference session so that session can be executed more efficiently when resumed.

[0170] 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.

[0171] 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.

[0172] 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) 2PCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0173] 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.

[0174] 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.

[0175] Model host 31 can access a library of pre-trained adapters or LoRA modules that can adapt a baseline model to align its outputs with a desired performance profile, augment model capabilities (e.g., to adapt to a different input modality, etc.), and the like. For instance, model host 31 can receive an input request to load a customized model, and model host 31 can retrieve one or more components to adapt a baseline model to the custom profile. Model host 31 can determine that a particular functionality is needed for a particular task (e.g., based on an output of a model that preprocesses an input) and retrieve a pre-trained component accordingly.

[0176] 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 dataPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0177] 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.

[0178] 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 canPCT / US24 / 60996 19 December 2024 (19.12.2024) 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).

[0179] 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.

[0180] 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.

[0181] 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 statisticalPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0182] 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.

[0183] 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.

[0184] 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0185] 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.

[0186] 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.

[0187] 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 textualPCT / US24 / 60996 19 December 2024 (19.12.2024) 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.

[0188] 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).

[0189] 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).

[0190] 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 toPCT / US24 / 60996 19 December 2024 (19.12.2024) 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).

[0191] Figure 11 is a block diagram of an example networked computing system that can perform aspects of example implementations of the present disclosure. The system can include a number of computing devices and systems that are communicatively coupled over a network 49. An example computing device 50 is described to provide an example of a computing device that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). An example server computing system 60 is described as an example of a server computing system that can perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Computing device 50 and server computing system(s) 60 can cooperatively interact (e.g., over network 49) to perform any aspect of the present disclosure (e.g., implementing model host 31, client(s) 32, or both). Model development platform system 70 is an example system that can host or serve model development platform(s) 12 for development of machine-learned models. Third-party system(s) 80 are example system(s) with which any of computing device 50, server computing system(s) 60, or model development platform system(s) 70 can interact in the performance of various aspects of the present disclosure (e.g., engaging third-party tools, accessing third-party databases or other resources, etc.).

[0192] Network 49 can be any type of communications network, such as a local area network (e.g., intranet), wide area network (e.g., Internet), or some combination thereof and can include any number of wired or wireless links. In general, communication over network 49 can be carried via any type of wired or wireless connection, using a wide variety of communication protocols (e.g., TCP / IP, HTTP, SMTP, FTP), encodings or formats (e.g., HTML, XML), or protection schemes (e.g., VPN, secure HTTP, SSL). Network 49 can also be implemented via a system bus. For instance, one or more devices or systems of Figure 11 can be co-located with, contained by, or otherwise integrated into one or more other devices or systems.

[0193] 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. ComputingPCT / US24 / 60996 19 December 2024 (19.12.2024) 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).

[0194] 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.

[0195] 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.

[0196] 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.

[0197] 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)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.

[0198] 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.

[0199] 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.

[0200] 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0201] 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.

[0202] 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).

[0203] Figure 11 illustrates one example arrangement of computing systems that can be used to implement the present disclosure. Other computing system configurations can be used as well. For example, in some implementations, one or both of computing system 50 or server computing system(s) 60 can implement all or a portion of the operations of model development platform system 70. For example, computing system 50 or server computing system(s) 60 can implement developer tool(s) 75 (or extensions thereof) to develop, update / train, or refine machine-learned models 1, 4, 16, 20, 55, 65, etc. using one or 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,PCT / US24 / 60996 19 December 2024 (19.12.2024) or refine machine-learned models based on local datasets (e.g., for model personalization / customization, as permitted by user data preference selections).

[0204] Figure 12 is a block diagram of an example computing device 98 that performs according to example embodiments of the present disclosure. Computing device 98 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 98 can include a number of applications (e.g., applications 1 through N). Each application can contain its own machine learning library and machine- learned model(s). For example, each application can include a machine-learned model. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. As illustrated in Figure 12, each application can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, each application can communicate with each device component using an API (e.g., a public API). In some implementations, the API used by each application is specific to that application.

[0205] Figure 13 is a block diagram of an example computing device 99 that performs according to example embodiments of the present disclosure. Computing device 99 can be the same as or different from computing device 98. Computing device 99 can be a user computing device or a server computing device (e.g., computing device 50, server computing system(s) 60, etc.). Computing device 98 can implement model host 31. For instance, computing device 99 can include a number of applications (e.g., applications 1 through N). Each application can be in communication with a central intelligence layer. Example applications include a text messaging application, an email application, a dictation application, a virtual keyboard application, a browser application, etc. In some implementations, each application can communicate with the central intelligence layer (and model(s) stored therein) using an API (e.g., a common API across all applications).

[0206] The central intelligence layer can include a number of machine-learned models. For example, as illustrated in Figure 13, a respective machine-learned model can be provided for each application and managed by the central intelligence layer. In other implementations, two or more applications can share a single machine-learned model. For example, in some implementations, the central intelligence layer can provide a single 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)

[0207] The central intelligence layer can communicate with a central device data layer. The central device data layer can be a centralized repository of data for computing device 99. As illustrated in Figure 13, the central device data layer can communicate with a number of other components of the computing device, such as, for example, one or more sensors, a context manager, a device state component, or additional components. In some implementations, the central device data layer can communicate with each device component using an API (e.g., a private API).

[0208] 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.

[0209] 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.

[0210] 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.PCT / US24 / 60996 19 December 2024 (19.12.2024)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.”

[0211] 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.

[0212] 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

PCT / US24 / 60996 19 December 2024 (19.12.2024)WHAT IS CLAIMED IS:

1. A computer-implemented method for tuning a text-to-image generation model, comprising: generating, by a computing system, training data by: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content item performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality data and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning the text-to-image model using the data in the preference data structure.

2. The computer-implemented method of claim 1, wherein determining a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data comprises: generating, by the computing system, a first preference value for the first image and a second preference value for the second image based on the image quality data and the content item performance signal data; and determining, by the computing system, the first image as the preferred image based on a comparison of the first preference value and the second preference value.

3. The computer-implemented method of claim 2, wherein the first preference value and the second preference value are generated by at least one of (i) a scalar model or (ii) a machine-learned model.PCT / US24 / 60996 19 December 2024 (19.12.2024)4. The computer-implemented method of claim 3, wherein the scalar model is configured to generate a preference value for an image based on a weighted combination of the image quality data and the content item performance signal data.

5. The computer-implemented method of any of claim 3 or 4, wherein the machine-learned model generates the first preference value and the second preference value based on combining the image quality data and the content item data for the respective images into a single embedding vector for each respective image.

6. The computer-implemented method of any preceding claim, wherein the image quality data comprises a plurality of instances of the first image, the method comprising: generating an average preference value for the first image based on the plurality of instances of the first image.

7. The computer-implemented method of any preceding claim, wherein the image quality data comprises a plurality of instances of the first image, the method comprising: generating a plurality of instances of the first image, wherein the preference data for each respective instance of the plurality of instances of the first image comprises context data.

8. The computer-implemented method of any preceding claim, wherein the image quality data comprises a quality score generated by a machine-learned model configured to generate a quality score for an image based on a plurality of attributes associated with the image.

9. The computer-implemented method of claim 8, wherein the plurality of attributes associated with the image comprise at least one of: a resolution of the image, a contrast of the image, a blurriness of the image, a brightness of the image, a saturation of the image, a hue of the image, a sharpness of the image, or a colorfulness of the image.PCT / US24 / 60996 19 December 2024 (19.12.2024)10. The computer-implemented method of claim 9, wherein the plurality of attributes associated with the image are generated by a machine-learned model configured to generate a quality score for an image based on the plurality of attributes associated with the image.

11. The computer-implemented method of any preceding claim, wherein the text- to-image model comprises a machine-learned model.

12. The computer-implemented method of any preceding claim, wherein the text- to-image model comprises a large language model (LLM).

13. A computing system comprising: one or more processors; and one or more computer-readable media storing instructions that are executable to cause the one or more processors to perform operations, the operations comprising: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content item performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality data and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning a text-to-image model using the data in the preference data structure.PCT / US24 / 60996 19 December 2024 (19.12.2024)14. The computing system of claim 13, wherein determining a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data comprises: generating, by the computing system, a first preference value for the first image and a second preference value for the second image based on the image quality data and the content item performance signal data; and determining, by the computing system, the first image as the preferred image based on a comparison of the first preference value and the second preference value.

15. The computing system of claim 14, wherein the first preference value and the second preference value are generated by at least one of (i) a scalar model or (ii) a machine- learned model.

16. The computing system of claim 15, wherein the scalar model is configured to generate a preference value for an image based on a weighted combination of the image quality data and the content item performance signal data.

17. The computing system of any of claim 15 or 16, wherein the machine-learned model generates the first preference value and the second preference value based on combining the image quality data and the content item data for the respective images into a single embedding vector for each respective image.

18. The computing system of any of claim 13 to 17, wherein the image quality data comprises a plurality of instances of the first image, the operations comprising: generating an average preference value for the first image based on the plurality of instances of the first image.

19. The computing system of any of claim 13 to 18, wherein the image quality data comprises a plurality of instances of the first image, the operations comprising: generating a plurality of instances of the first image, wherein the preference data for each respective instance of the plurality of instances of the first image comprises context data.PCT / US24 / 60996 19 December 2024 (19.12.2024)20. One or more transitory or non-transitory computer readable media storing instructions that are executable by one or more processors to perform operations comprising: generating, by a computing system, training data by: accessing, by the computing system, a plurality of images associated with a plurality of content items; accessing, by the computing system, for each image of the plurality of images, image quality data; accessing, by the computing system, a plurality of content item performance signal data associated with the plurality of content items; generating, by the computing system, preference data from the content item performance signals by: selecting, by the computing system, a first image and a second image based on the image quality data and content item performance signal data; determining, by the computing system, a preferred image of the first image and the second image based on analysis of the image quality data and the content item performance signal data; and storing, by the computing system, the selected first image and second image and the preferred image in a preference data structure; and tuning a text-to-image model using the data in the preference data structure.

Images