I’m using a personal account. I’m running a custom-handler Inference Endpoint for text embeddings (Llama-3.1-8B). Since early this morning on 2026-07-14, I’m getting intermittent HTTP 500s (~70% of requests fail, ~30% return 200 normally).
What I’ve ruled out:
Not GPU-specific - happens on both L40S and A100.
Not endpoint-specific - brand-new endpoints show the same. Not my code/deps - the same build ran hundreds of thousands of requests fine yesterday.
Container is healthy - model loads successfully, and successful requests return correct results in 30–415 ms.
Key finding: the container logs only show the successful requests. The failed (500) requests don’t appear in the container logs at all — they seem to be failing at the gateway/router layer before reaching the container.
The status page shows no incident. Is there a known issue with the us-east-1 inference gateway, or something on my account causing this? Happy to share endpoint IDs privately.
Thanks!