mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-14 23:21:35 +00:00
e118ebec55
Team BYOK deployments share their public model name with admin-managed models, so billing pulled from /global/spend/report cannot distinguish spend a team already paid its provider for, and that spend gets billed twice. Zeroing the deployment cost is not a workaround because spend drives member-level budget enforcement. When exclude_team_models=true, every SQL branch of the report anti-joins LiteLLM_SpendLogs.model_id against LiteLLM_ProxyModelTable rows whose model_info carries a team_id. Ownership via model_info.team_id (the same signal Router._is_team_specific_model reads) survives deployment renames and never matches admin models, unlike the mangled model_name prefix. With the flag absent or false every query is byte-identical to before. Known limitation: spend from deployments deleted since the request was logged cannot be classified and remains included. |
||
|---|---|---|
| .. | ||
| litellm-dashboard | ||
| Dockerfile | ||
| nginx.conf | ||