Skip to main content
huggingface.health
Inference Endpoints

Is Hugging Face Inference Endpoints down? Live status

Dedicated GPU inference on AWS/Azure/GCP

Live status for Hugging Face Inference Endpoints — the dedicated GPU deployment control-plane at api.endpoints.huggingface.cloud.

Hugging Face Inference Endpoints provides dedicated, autoscaling infrastructure for deploying transformers and LLMs on secure cloud GPUs and CPUs. You can monitor live uptime, regional latency, and system outages at huggingface.health to verify endpoint availability. Inference Endpoints deploy dedicated ML models on fully managed GPU and CPU instances — check live service status, control-plane availability at api.endpoints.huggingface.cloud, and current operational latency directly via real-time client-side health probes.

Inference Endpoints API Status

Live, in-browser checks against Inference Endpoints's public APIs. Refreshes every 15s.

Checking…

Control plane

Endpoints API and dashboard

2 endpoints
  • Dashboard
    ui.endpoints.huggingface.cloud
    —ms
    warming up
  • Endpoints API
    api.endpoints.huggingface.cloud
    —ms
    warming up
Polling every 15s · Last updated — · Next in 15sOpen Inference Endpoints ↗

Individual endpoint URLs are per-tenant and can't be probed from a public dashboard. These checks confirm reachability of the shared control plane.

Sponsored · Inference Endpoints
Ask Endo
Dedicated · GPU

I'm Endo — your Inference Endpoints guide. Ask about GPU SKUs, autoscaling, or cold starts.

Open Endpoints

Sponsored content. huggingface.health may earn a referral reward when you sign up via this widget. Answers are informational only — not professional advice.

About Inference Endpoints

Inference Endpoints let you deploy any model from the Hub onto dedicated GPU or CPU instances across AWS, Azure, and GCP — with autoscaling, scale-to-zero, and per-second billing. This page tracks the shared control plane.

Inference Endpoints at a glance

Clouds
AWS, Azure, GCP
Billing
Per-second, per-instance
Scale-to-zero
Optional idle sleep
Model source
Any Hub model
Best for
Production, predictable latency

Common Inference Endpoints issues

  • Cold-start on first request

    Scale-to-zero endpoints must load model weights before serving. Large LLMs take 60–180s. Turn off scale-to-zero for latency-sensitive traffic.

  • GPU capacity error at deploy

    Some GPU SKUs (A100, H100) have limited regional capacity. Try a different region or downshift to L4/A10G if latency budget allows.

  • Quota exceeded on the org

    Orgs have vCPU/GPU quotas. Request an increase from the Endpoints dashboard — approvals for standard SKUs are usually fast.

Background

Hugging Face launched Inference Endpoints in 2022 to give teams a one-click path from a Hub model card to a production-ready HTTPS endpoint. It sits alongside the Serverless Inference API and the newer OpenAI-compatible router as the paid, dedicated tier.

Frequently asked

How do I check if Hugging Face Inference Endpoints are down?
Check real-time status and latency metrics on huggingface.health or test endpoint reachability via cURL. Hugging Face also provides incident logs and regional availability updates across AWS, GCP, and Azure infrastructure.
What causes Hugging Face dedicated inference endpoint failures?
Common causes include GPU quota limits, cold boot initialization timeouts, container memory exhaustion (OOM), cloud provider regional disruptions, or unhandled invalid model weights in the inference container.
Can Hugging Face Inference Endpoints scale to zero?
Yes, Inference Endpoints support scale-to-zero to eliminate idle compute costs. When a new request arrives, the endpoint automatically spins up the GPU container, incurring a brief cold start latency.
What causes Hugging Face dedicated inference endpoint errors?
Common errors stem from GPU initialization cold starts, out-of-memory (OOM) failures with large weights, reaching concurrency limits, or temporary regional outages across underlying cloud providers (AWS or Azure).
How do Hugging Face Inference Endpoints differ from Serverless Inference?
Inference Endpoints run dedicated container instances with custom autoscaling (including scale-to-zero) and guaranteed GPUs, whereas Serverless Inference shares shared infrastructure with stricter rate limits and shared model caches.
How can I check if Hugging Face Inference Endpoints is down?
Check huggingface.health/inference-endpoints for live browser-based latency checks, or send a ping request directly to api.endpoints.huggingface.cloud using curl to verify API availability and authentication.
Why is my Hugging Face Inference Endpoint timing out?
Timeouts typically occur during model cold starts, replica scaling, GPU provisioning delays on AWS or Azure, or when hitting concurrent request limits on single-instance endpoints.
What is the uptime SLA for Hugging Face Inference Endpoints?
Dedicated Inference Endpoints run on enterprise cloud infrastructure with high availability SLAs determined by your underlying cloud provider (AWS, Azure, GCP) and Hugging Face Enterprise tier agreement.

About this check

All data is provided on a best-effort basis with no guarantees about accuracy or availability. This is an unofficial status page and is not affiliated with, endorsed by, or sponsored by Hugging Face, Inc.

Checks run client-side from your browser against Inference Endpoints's public endpoints. Latency reflects your network path, not the server's health alone. See our resources for more on how to read these checks.

Related on this site