Docs / Model runtime
Client compatibility and loading feedback
Let clients discover models and handle loading responses instead of treating a swap as a failure.
Clients should read /v1/models before choosing a model and be prepared for a
request to wait while llama-swap loads it. Set sendLoadingState: true when a
client can display a loading response rather than timing out or retrying
blindly. includeAliasesInList also exposes configured aliases in model lists.
sendLoadingState: true
includeAliasesInList: true
This changes feedback, not loading speed. Keep the client timeout longer than the model's startup time and use a client that understands the selected API.