DYNAMICALLY SELECTING ARTIFICIAL INTELLIGENCE MODELS AND HARDWARE ENVIRONMENTS TO EXECUTE TASKS
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.
1 . A computer-implemented method comprising:
monitoring hardware usage metrics for a hardware environment hosting a machine-learning model;
extracting, at a content management system, workload data requesting execution of a training task using the machine-learning model;
based on the workload data, initiating the training task at the hardware environment based on detecting a bandwidth availability indicated by the hardware usage metrics; and
pausing the training task based on detecting a change in the bandwidth availability indicated by the hardware usage metrics.
2 . The computer-implemented method of claim 1 , wherein initiating the training task at the hardware environment further comprises:
identifying that the hardware usage metrics indicate that the bandwidth availability satisfies a training task bandwidth usage threshold; and
initiating the training task based on identifying that the bandwidth availability satisfies the training task bandwidth usage threshold.
3 . The computer-implemented method of claim 1 , further comprising:
forecasting a minimum-use time period where an average bandwidth availability satisfies a training task bandwidth usage threshold; and
scheduling the training task for the minimum-use time period.
4 . The computer-implemented method of claim 3 , further comprising initiating the training task at the hardware environment during the minimum-use time period based on detecting that an actual bandwidth usage at the minimum-use time period satisfies a training task bandwidth usage threshold.
5 . The computer-implemented method of claim 1 , wherein initiating the training task at the hardware environment further comprises:
monitoring hardware usage metrics by monitoring a user count metric, a memory usage metric, a central processing unit metric, or a graphics processing unit metric; and
initiating the training task at the hardware environment based on two or more of the user count metric, the memory usage metric, the central processing unit metric, and the graphics processing unit metric.
6 . The computer-implemented method of claim 1 , wherein pausing the training task further comprises:
monitoring the hardware usage metrics for the hardware environment throughout a use time period to identify a high-use time period; and
pausing the training task based on forecasting the change in the bandwidth availability at the high-use time period.
7 . The computer-implemented method of claim 1 , wherein extracting workload data requesting execution of the training task further comprises:
determining an estimated processing requirement and an estimated storage requirement for executing the training task; and
determining an estimated execution length for the training task.
8 . The computer-implemented method of claim 1 , further comprising:
receiving the workload data by receiving an indication that the training task does not require continuous execution; and
initiating the training task based on the indication that that the training task does not require continuous execution.
9 . The computer-implemented method of claim 1 , further comprising:
initiating the training task based on identifying that the hardware usage metrics indicate that the bandwidth availability satisfies a training task bandwidth usage threshold; and
pausing the training task based on identifying that the hardware usage metrics indicate that the bandwidth availability no longer satisfies the training task bandwidth usage threshold.
10 . A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computer system to:
monitor hardware usage metrics for a hardware environment hosting a machine-learning model;
extract, at a content management system, workload data requesting execution of a batch task using the machine-learning model;
based on the workload data, initiate the batch task at the hardware environment based on detecting a bandwidth availability indicated by the hardware usage metrics; and
pause the batch task based on detecting a change in the bandwidth availability indicated by the hardware usage metrics.
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:
forecast a minimum-use time period where an average bandwidth availability satisfies a batch task bandwidth usage threshold; and
schedule the batch task for the minimum-use time period.
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 resume the batch task based on detecting an additional change in the bandwidth availability.
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 extract workload data by:
determining an estimated processing requirement and an estimated storage requirement for executing the batch task; and
determining an estimated execution length for the batch task.
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 initiate the batch task at the hardware environment by:
identifying that the hardware usage metrics indicate that the bandwidth availability satisfies a batch task bandwidth usage threshold; and
initiating the batch task based on identifying that the bandwidth availability satisfies the batch task bandwidth usage threshold.
15 . The non-transitory computer-readable medium of claim 14 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:
determine that the bandwidth availability indicated by the hardware usage metrics no longer satisfies the batch task bandwidth usage threshold; and
pause the batch task based on determining that the bandwidth availability no longer satisfies the batch task bandwidth usage threshold.
16 . A system comprising:
at least one processor; and
at least one non-transitory computer-readable storage medium storing instructions that, when executed by the at least one processor, cause the system to:
monitor hardware usage metrics for a hardware environment hosting a machine-learning model;
extract, at a content management system, workload data requesting execution of a training task using the machine-learning model;
determine, at the content management system, that a training dataset for the training task is compatible with the hardware environment; and
based on determining that the training dataset is compatible with the hardware environment and the workload data, initiate the training task at the hardware environment based on a bandwidth availability indicated by the hardware usage metrics.
17 . The system of claim 16 , further comprising instructions that, when executed by the at least one processor, cause the system to determine that the training dataset is compatible with the hardware environment by identifying that the training dataset is local to the hardware environment.
18 . The system of claim 16 , further comprising instructions that, when executed by the at least one processor, cause the system to initiate the training task by:
determining that the training dataset is compatible with the hardware environment based on determining a data affinity for the hardware environment and the training dataset; and
initiating the training task based on determining the data affinity for the hardware environment and the training dataset.
19 . The system of claim 16 , further comprising instructions that, when executed by the at least one processor, cause the system to:
forecast a minimum-use time period where the hardware usage metrics indicate that an average bandwidth availability satisfies a training task bandwidth usage threshold;
schedule the training task for the minimum-use time period; and
initiate the training task based on detecting that an actual bandwidth at the minimum-use time period satisfies the training task bandwidth usage threshold.
20 . The system of claim 16 , further comprising instructions that, when executed by the at least one processor, cause the system to:
pause the training task based on detecting a change in the bandwidth availability indicated by the hardware usage metrics; and
resume the training task based on detecting an additional change in the bandwidth availability indicated by the hardware usage metrics.