BRK246 Breakout Wed 18 Nov 13:45 PST
From open weights to workload: Operationalizing at enterprise scale
Manoj Bableshwar, Gulsimo Osimi, Miguel Armenta, Pavan Tagirisa
Summary
From the recording
Microsoft Foundry managed compute, for deploying open source models on dedicated GPUs with Microsoft running the runtime and patching, was announced as generally available, billed per GPU hour. Speakers said more than 150 models shipped in the last three to four months and a data zone option covers US or EU. Programmatic training and LoRA adapter deployment (up to 8 adapters per deployment) were described as preview. Caveat: no pricing was given, and UK or EU-UK data zone availability was not stated.
Session description
Microsoft's catalogue description
Open-weight models offer greater choice and control, but production at scale requires the right infrastructure, serving stack, and operations. See how Microsoft Foundry Managed Compute helps teams move from model selection to enterprise deployment. Learn from AT&T’s OTel 2.0 experience using multiple open models and GPU types to balance performance, economics, and workload needs without standardizing on one model.