Details
-
Feature
-
Could have
-
None
-
SRCnet
-
-
-
0.5
-
1
-
0
-
-
-
-
example-workflows-and-benchmarks tests-compilation
Description
Wandb.ai provides a remote server that tracks metrics when a task/workflow is run, allowing us to tune parameters and refine how hardware is used (e.g. CPU/GPU/RAM use). It is particularly designed for machine learning applications, allowing you to keep track of your models, loss functions and metrics for any run. In our case, we can make use of it for the ML task, but it will provide benefits to all the tasks in the repo.
It is free for academic institutions and personal accounts.
We also want to compare this to other metric tracking systems which can provide a much more complete picture of the compute load without bloat - this includes testing things like Nsight and Intel Advisor, and collecting the metric tracking information in a consistent way.