Troubleshooting model loading
When Bridge connects but the model won’t answer — or answers at a crawl.
Updated August 2026
Check in this order
- Is the runtime running? Bridge talks to your model server; if Ollama or vLLM is not up, there is nothing to talk to.
- Is the endpoint right? The URL in Bridge’s model settings must match where the runtime actually listens.
- Is the model present? A runtime can be healthy but not have the model pulled/downloaded yet.
- Does it fit in memory? Out-of-memory failures and silent CPU fallback are the two classic large-model symptoms.
The slow-but-working case
If answers come but take forever, the model is likely running on CPU instead of GPU. Check the runtime’s own logs or status output — it will say what device the model loaded on.
Still stuck
The runtime’s logs nearly always name the real problem. Bring that log line with you when you ask the team for help.