Hi Hailo team! I tried running parallel hailortcli run2
benchmarks using the –multi-process-service
flag. For the run2
command, this flag doesn’t seem to work; while for run
it works as expected.
Hey @Nils-Oliver!
I can see the confusion here. The --multi-process-service
flag is designed specifically for hailortcli run
and won’t work with run2
- that’s why you’re seeing it function properly with the regular run
command but getting ignored when using run2
.
Here’s the deal: you’ve got two different approaches that don’t play together:
If you need multi-process sharing: Use hailortcli run
with the --multi-process-service
flag. This creates a service that lets multiple processes share the device without conflicts.
If you need run2
features (like the scheduler or multi-network fairness): You’ll have to work within a single process. Load all your networks in one run2
session and let the Model Scheduler handle everything internally, or use the C++/Python scheduler APIs.
Bottom line: it’s one or the other - you can’t combine the multi-process service with run2
’s scheduler features. Choose based on what’s more critical for your specific use case.
Hope this clears things up!