Eero Tamminen

@vrantala Rebased to `main` and fixed the conflicts => file read + `b64encode()` (for video/audio processing) is now also done with `async`/`await`.

> Do you think we would be able to use a heuristic of checking CPU capabilities as a close-enough deciding factor? Maybe check also number of _available_ CPU cores? Single...

> on an example system I see the dGPU listed as `card1` while the iGPU is listed as `card0` Device name indexes and device file names in general, come from...

> Denylist has the issue with cards enumerating in different order. > ... > In general, I don't think one should depend on having card0 to be iGPU or the...

With #2101 one can tell GPU plugin to accept or deny GPUs having specific IDs. What if it would be possible to run multiple GPU plugin set instances with different...

Docker supports passing username/groupname, but it reads them from inside the container, not from host which IMHO makes that feature useless (only reason I can think of for using names...

@cyphar > It's not useless -- how else would you run a program as a non-root user that is configured inside the container (note: this is the mechanism by which...

> Prometheus-Adapter can be seen as Prometheus-related component, but I wouldn't say its essential for a Prometheus-based monitoring stack. At this point in time, I'd be against adding extra complexity...

Newer PR https://github.com/opea-project/GenAIExamples/pull/2012 increases timeouts but seems otherwise same, it just has much less descriptive title. Maybe the other one of these could be closed?

I haven't used that plugin, but maybe you could install e.g. Prometheus to query & store metrics, and Grafana to visualize them? (They prefer total metrics, as then user is...