mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-24 00:52:24 +00:00
30acb09600
The "Edit Model Settings" form re-sent the entire litellm_params on every save, so the backend re-encrypted every value; a masked secret from /model/info would be re-encrypted over the real one, silently corrupting credentials, and stale values were resurrected Frontend: seed the form once, then PATCH only the fields the user actually changed (touch-gated scalars, per-key diff of the LiteLLM Params JSON), send model_info only when a contributing field changed while re-injecting the real id/db_model, and make the masked-secret guard recurse so a secret nested inside an object such as extra_headers.Authorization can't slip through; when a redacted value is dropped the user is warned instead of it failing silently Backend: update_db_model merges via model_dump(exclude_unset=True) filtered to non-None values, so a minimal PATCH preserves every field the client did not send; this fixes the four default-False booleans (use_in_pass_through, use_litellm_proxy, use_xai_oauth, merge_reasoning_content_in_choices) that exclude_none re-sent as False and clobbered stored True values |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||