[KongchangAI]
Conceptmodel workload

Model Workload

A concrete task instance that occupies computing resources, such as training, fine-tuning, inference, or evaluation, and that may exist as a job, a container, or a distributed process. In large clusters, a single model often corresponds to multiple parallel tasks, so the workload is better suited as the object of tracking and control.

Timeline (last 90 days)

Nov 9

Control Room's core capabilities include identifying and tracking model workloads across research infrastructure, providing global visibility.

Unverified50%
Nov 9

In large research clusters, the same model often corresponds to multiple parallel jobs, and jobs start and stop dynamically as experiments progress.

Unverified50%
Nov 9

A model workload refers to the sum of the computing resources and running tasks consumed by a model and its accompanying tasks during a single training or inference process.

Unverified70%
Nov 9

At large scale, a single model workload often spans hundreds or thousands of accelerator cards and is distributed across the cluster in the form of jobs or container orchestration units.

Unverified50%

All Facts (4)

Source Articles