FlowtandBriefAI · Software · ResearchMeet Flowtand

Hugging Face and Microsoft Expand Open-Model Collaboration Across Azure and VS Code

Announcement

Hugging Face

Browse Industry topic page →

Hugging Face announced at Microsoft Build 2024 a broader collaboration with Microsoft: popular open LLMs gain 1-click deployment from the Azure AI Studio Hugging Face Collection, AMD MI300X VMs reach general availability on Azure with Optimum-AMD support, Microsoft's Phi-3 models stay permissively licensed, and a beta Spaces Dev Mode links Hugging Face Spaces to VS Code for PRO subscribers.

What Hugging Face and Microsoft announced

In a May 21, 2024 post tied to Microsoft Build, Hugging Face described a deepened collaboration with Microsoft spanning cloud, open science, open source and developer tooling. The concrete change on the cloud side is an expanded Hugging Face Collection in the Azure Model Catalog: Hugging Face says popular open models including Meta's Llama 3, Mistral 7B, Cohere's Command R Plus, Qwen 1.5 110B and highly ranked community fine-tunes from the Open LLM Leaderboard can now be deployed with one click from Azure AI Studio, or started from a model card on the Hub via a "Deploy on Azure" option. Separately, Microsoft made Azure ND MI300X virtual machines based on AMD Instinct MI300 GPUs generally available; Hugging Face says it worked with AMD and Microsoft on optimization through its Optimum-AMD library. The post also notes Microsoft's roughly 300 models on the Hub, including Phi-3 under an MIT license, and the beta Spaces Dev Mode for PRO subscribers, which connects a Space to local or web-hosted VS Code.

A scenario for a platform engineer

Consider a platform engineer at a company that already runs workloads in Azure and wants to test open models without building bespoke deployment pipelines. A team could try using the Hugging Face Collection in Azure AI Studio to stand up one candidate model in its own Azure account via the deploy action, then compare operational effort with its existing process. The MI300X VMs and Optimum-AMD tooling are a second thread a performance-focused team could investigate if GPU cost or throughput is a concern, treating Hugging Face's stated optimization work as a lead rather than a result. For teams building demo apps, Spaces Dev Mode might be worth a look if they already hold PRO subscriptions and want to edit a Space in a desktop editor.

How to trial it and measure

A reasonable first experiment is narrow: pick one model from the collection, deploy it in a non-production Azure subscription, and record time-to-first-inference, setup steps and configuration effort against the team's current baseline. Success could be defined as fewer manual steps or shorter setup time, measured by the same person on both paths. For GPU work, run a representative workload on the new VM class versus the existing baseline and log throughput and cost, but treat any figures as your own measurements, not confirmation of vendor claims. Failure checks should include verifying the deployed model actually matches the model card and that quotas and regional availability hold. Keep a human reviewer in the loop before promoting anything, and note that Spaces Dev Mode is beta, so avoid depending on it for critical workflows.

Trade-offs and when to hold back

The convenience of one-click deployment trades away some control: teams used to custom pipelines may find the managed path constraining, and the post does not detail governance, pricing or data-handling terms, so those need separate verification. The AMD VM option requires specific hardware and tooling knowledge, and Optimum-AMD support should be validated on the exact models you need. Spaces Dev Mode's beta status and PRO-only availability make it unsuitable for production dependencies today. Where the evidence is thin — vendor performance comparisons, model quality rankings — teams should not treat the announcement as proof. If your organization needs audited deployment paths or cannot use Azure, this collaboration's headline features may not apply.

Context and limitations

  • Based on a short source excerpt; the full article was not analyzed.
  • Spaces Dev Mode is described as beta and limited to PRO subscribers.
  • Performance claims, such as Phi-3 mini's comparison to larger models, come from the publisher and were not independently tested here.
  • Based on bounded source text, not independent testing. Practical suggestions are editorial proposals.