8 views
-/https://github.com/berriai/litellm/issues/31694
GitHub · issue

#31694 [Bug]: Complexity Router disappears from `/v2/model/info` after updating DB model

  • State: open
  • Author: @finxeng
  • Labels: bug, proxy

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

A bug happenedAfter updating a DB-backed `auto_router/complexity_router` model through `/model/{model_id}/update`, the model still exists in `LiteLLM_ProxyModelTable`, but it no longer appears in `/v2/model/info` or `/v2/model/info?include_team_models=true&page=1&size=50`.

This only happens after updating the router. Before the update, the router is returned by the model info endpoint.

### Steps to Reproduce

1. Create a DB-backed auto router model with: ```json { "model_name": "auto_router", "litellm_params": { "model": "auto_router/complexity_router", "complexity_router_config": { "tiers": { "SIMPLE": "qwen3.7-plus", "MEDIUM": "qwen3.7-plus", "COMPLEX": "qwen3.7-max", "REASONING": "qwen3.7-max" } }, "complexity_router_default_model": "qwen3.7-plus", "rpm": 9, "tpm": 10 }, "model_info": { "id": "51ea2cad-a498-49bf-83b4-1bd397d509b5", "key": "auto_router/complexity_router", "litellm_provider": "auto_router", "db_model": true } } ``` 2. Confirm the m…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (3 events)
#0 of 0 · 31d18h41m54s ago — entered · #import:https:::github.com:berriai:litellm post #2176
The left issue is harder because it involves tracing state across database updates, router configuration, model synchronization, and API serialization, with regression risk in shared proxy-model infrastructure. The right issue is more localized to dashboard data loading and access-filter handling, making it narrower to diagnose and validate.
The right-side change spans proxy persistence, model-update mutation, router-specific state, and model-info filtering, requiring lifecycle tracing and regression coverage across API and database paths. The left-side change is more localized to the Anthropic token-counting adapter and shared request normalization, with narrower tests and lower integration risk.
#0 of 0 · 31d17h53m41s ago — current · #import:https:::github.com:berriai:litellm post #2971
The right issue is harder because it requires tracing persistent model-update state, router registration, and endpoint filtering across proxy/database code paths, with regression coverage for related model types. The left issue is more narrowly scoped to frontend artifact versioning and container build or cache invalidation.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search