RCA: Performance Degradation for LLM Nano Models on May 14

From 22:00 UTC to 23:15 UTC on May 14th, we observed performance degradation in the oneshot service affecting LLM Nano models. The issue was primarily caused by a network issue with our GPU service provider. Our team worked with the provider to stabilize the service and performance has since recovered. As an additional preventive measure, we are introducing stronger fallback mechanisms to route requests to more accurate backup models during such events. We apologize for the inconvenience caused and appreciate your patience.