BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.devcon.org//devcon8//talk//HRSPVY
BEGIN:VTIMEZONE
TZID:IST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:IST
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devcon8-HRSPVY@cfp.devcon.org
DTSTART;TZID=IST:20261105T161500
DTEND;TZID=IST:20261105T162000
DESCRIPTION:Open inference engines give you many knobs and it’s easy to t
 urn the wrong ones. In this 5-minute lightning talk\, I’ll share one con
 crete lesson from tuning an open-source model server for better real-world
  performance: what we changed\, what metric we watched\, and what surprise
 d us.\n\nThe goal isn’t a full benchmark suite. It’s a tight takeaway 
 you can reuse when comparing open serving stacks or tuning memory\, concur
 rency\, and startup behavior on GPU hosts. Come for one practical lesson\;
  leave with a sharper instinct for which knobs matter first.
DTSTAMP:20261101T123115Z
LOCATION:Community Led Stage
SUMMARY:Stop twisting every dial: One open Inference tuning lesson - Aqsa A
 qeel
URL:https://cfp.devcon.org/devcon8/talk/HRSPVY/
END:VEVENT
END:VCALENDAR