Add vLLM models authentication diagnostics

This commit is contained in:
2026-11-22 18:17:30 +00:00
parent ca8cf1b827
commit c3257b63f0
9 changed files with 116 additions and 9 deletions

View File

@@ -73,6 +73,15 @@ not validation of the full account document. Other statuses and unrecognized
recognition allowance reports `authentication_response_limit_exceeded`, leaving
credential acceptance unknown; see the [public reason](../../smoke_diagnostics.go).
The `vllm_models` adapter issues one `GET` to `/models` relative to the
configured base unless an explicit diagnostic endpoint is supplied. It sends
an available credential but does not require one. A 200 response is recognized
when it is a JSON object with `object: "list"` and an array-valued `data`;
empty arrays are accepted. A 401 means rejection; other responses do not
establish acceptance. Recognition uses the same bounded response handling as
the key probe. Acceptance does not establish that authentication is enforced or
that inference is authorized.
Diagnostic HTTP evidence describes the last request observed at Promptkit's
RoundTripper boundary and any response associated with it. It includes the
selected destination without URL user information, response status, and