The networking issue with our provider was resolved at 0940 UTC, and all requests have been running normally since then.

This appears to be an issue with the provider for our H100 and L40S cluster. We're working with our provider to resolve it.

Otherware hardware types are unaffected.

Requests for predictions made on models and deployments running on H100 and L40S instances are taking a long time to respond and sometimes timing out.

We're trying to establish the cause of the issues.