Dynamically selecting artificial intelligence models and hardware environments to execute tasks

Patent metadata

Field	Value
Publication number	US-2025238333-A1
Application number	US-202418732320-A
Country	US
Kind code	A1
Filing date	Jun 3, 2024
Priority date	Jan 22, 2024
Publication date	Jul 24, 2025
Grant date	—

Abstract

The present disclosure relates to systems, non-transitory computer-readable media, and methods for selecting machine-learning models and hardware environments for executing a task. In particular, in one or more embodiments, the disclosed systems select a designated machine-learning model for executing a task based on workload features of the task and task routing metrics for a plurality of machine-learning models. In addition, in one or more embodiments, the disclosed systems select a designated hardware environment for executing the task based on workload features for the task and task routing metrics for a plurality of hardware environments. In some embodiments, the disclosed systems select a fallback machine-learning model and a fallback hardware environment for executing the task if the designated machine-learning model or designated hardware environment are unavailable. Moreover, in one or more embodiments, the disclosed systems can pause and initiate tasks based on bandwidth availability.

First claim

1 . A computer-implemented method comprising:
receiving, from a device connected by a network to a content management system, workload data requesting execution of a task using a machine-learning model; extracting, from the workload data, workload features defining characteristics of the task; selecting, based on the workload features, a primary machine-learning model for executing the task and a fallback machine-learning model for executing the task if the primary machine-learning model is unavailable; and based on detecting that the primary machine-learning model is unavailable prior to executing the task, providing the workload data to a computing environment of the fallback machine-learning model for executing the task.

2 . The computer-implemented method of claim 1 , wherein extracting workload features defining characteristics of the task further comprises determining an estimated processing requirement and an estimated storage requirement for executing the task.

3 . The computer-implemented method of claim 1 , wherein selecting the primary machine-learning model and the fallback machine-learning model further comprises:
generating optimization metrics for each machine-learning model of a plurality of machine-learning models; and selecting the primary machine-learning model and the fallback machine-learning model based on the optimization metrics.

4 . The computer-implemented method of claim 1 , wherein selecting the primary machine-learning model and the fallback machine-learning model further comprises:
determining a model state for a plurality of machine-learning models; selecting the primary machine-learning model based in part on a first model state of the primary machine-learning model; and selecting the fallback machine-learning model based in part on a second model state of the fallback machine-learning model.

5 . The computer-implemented method of claim 1 , wherein selecting the primary machine-learning model and the fallback machine-learning model further comprises:
determining, based on the workload features, a financial cost metric, an execution time metric, an execution cost metric, or a model fit metric for executing the task on each machine-learning models in a plurality of machine-learning models; and selecting the primary machine-learning model and the fallback machine-learning model is based on two or more of the financial cost metric, the execution time metric, the execution cost metric, or the model fit metric.

6 . The computer-implemented method of claim 1 , further comprising:
selecting an additional machine-learning model for executing the task based on the workload features; and based on detecting that the fallback machine-learning model is unavailable, providing the workload data to a computing environment of the additional machine-learning model for executing the task.

7 . The computer-implemented method of claim 1 , further comprising selecting a trained machine-learning model as the primary machine-learning model and a third-party trained machine-learning model as the fallback machine-learning model.

8 . The computer-implemented method of claim 1 , further comprising:
detecting that the primary machine-learning model is unavailable based on identifying that a first hardware environment associated with the primary machine-learning model is unavailable to execute the task; and providing the workload data to a second hardware environment associated with the fallback machine-learning model based on identifying that the second hardware environment is available to execute the task.

9 . The computer-implemented method of claim 1 , wherein selecting the primary machine-learning model and the fallback machine-learning model further comprises:
performing a software domain analysis of each machine-learning model of a plurality of machine-learning models for executing the task; and selecting the primary machine-learning model and the fallback machine-learning model based on the software domain analysis.

10 . A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computer system to:
receive, from a device connected by a network to a content management system, workload data requesting execution of a task using a machine-learning model; extract, from the workload data, workload features defining characteristics of the task; determine task routing metrics for a plurality of machine-learning models hosted in respective network environments; select, based on the workload features and the task routing metrics, a primary machine-learning model for executing the task and a fallback machine-learning model from the plurality of machine-learning models for executing the task if the primary machine-learning model is unavailable; and based on detecting that the primary machine-learning model is unavailable prior to executing the task, provide the workload data to a computing environment of the fallback machine-learning model for executing the task.

11 . The non-transitory computer-readable medium of claim 10 , further comprising instructions that, when executed by the at least one processor, cause the computer system to select the primary machine-learning model and the fallback machine-learning model by:
determining task routing metrics for the plurality of machine-learning models by determining a financial cost metric, an execution time metric, an execution cost metric, or a model fit metric for executing the task on each machine-learning models of the plurality of machine-learning models; and selecting, from the plurality of machine-learning models, the primary machine-learning model and the fallback machine-learning model based on two or more of the financial cost metric, the execution time metric, the execution cost metric, or the model fit metric for executing the task.

12 . The non-transitory computer-readable medium of claim 10 , further comprising instructions that, when executed by the at least one processor, cause the computer system to select the primary machine-learning model and the fallback machine-learning model by:
generating, based on the workload data and the task routing metrics, an optimization metric for each machine-learning model of the plurality of machine-learning models; and selecting the primary machine-learning model and the fallback machine-learning model from the plurality of machine-learning models based on a first optimization metric for the primary machine-learning model and a second optimization metric for the fallback machine-learning model.

13 . The non-transitory computer-readable medium of claim 10 , further comprising instructions that, when executed by the at least one processor, cause the computer system to select the primary machine-learning model and the fallback machine-learning model by:
accessing historical user feedback data about executing tasks using one or more machine-learning models of the plurality of machine-learning models; determining a historical quality metric for each machine-learning model of the plurality of machine-learning models based on the historical user feedback data; and selecting the primary machine-learning model and the fallback machine-learning model from the plurality of machine-learning models based in part on a first historical quality metric for the primary machine-learning model and a second historical quality metric for the fallback machine-learning model.

14 . The non-transitory computer-readable medium of claim 10 , further comprising instructions that, when executed by the at least one processor, cause the computer system to select the pr

Assignees

• Dropbox Inc

Inventors

• Poothiyot Ashok Pancily
• Zafar Ali
• Penta Anthony
• Voorhees Stephen
• Gasser Tim
• Chang Tsung-Hsiang
• Hulten Geoff

Classifications

• G06F9/485

  Task life-cycle, e.g. stopping, restarting, resuming execution (G06F9/4881 takes precedence) · CPC title

• G06F9/5011

  the resources being hardware resources other than CPUs, Servers and Terminals · CPC title

• H04L43/0894

  Packet rate · CPC title

• H04L41/16 Primary

  using machine learning or artificial intelligence · CPC title

• G06F2209/509

  Offload · CPC title