[Bug]: /rerank does not apply Virtual Key fallback chain (Fallbacks=None)

[Sunsilkk]

Check for existing issues

• I have searched the existing issues and checked that this is not a duplicate.
• I found a rerank issue (Facing the Issue with the rerank model of cohere.rerank-v3-5:0 #20473), but it is about provider auth for Cohere rerank, not Virtual Key fallback handling on /rerank.

What happened?

When configuring a Virtual Key in the LiteLLM UI with a Fallback Chain for a rerank model, requests to /rerank do not see any fallback configuration.

In the Virtual Keys UI:

• Primary model: Qwen3-Reranker-8B
• Fallback chain includes: deepinfra/Qwen/Qwen3-Reranker-8B

Then I create/copy the virtual key and call /rerank with:

• model: "Qwen3-Reranker-8B"
• mock_testing_fallbacks: true

Expected behavior:

• LiteLLM should pick up the fallback chain configured on the key
• mock_testing_fallbacks: true should trigger fallback routing
• Request should continue to the fallback model instead of failing with Fallbacks=None

Actual behavior:

• LiteLLM throws an internal mock exception and reports:
  • Fallbacks=None
  • Available Model Group Fallbacks=None

This makes it look like the fallback chain saved on the Virtual Key is not being applied to /rerank requests at all.

Steps to Reproduce

1. Open LiteLLM UI -> Virtual Keys
2. Create a new virtual key
3. Set primary model to Qwen3-Reranker-8B
4. In Router Settings, configure a fallback chain with deepinfra/Qwen/Qwen3-Reranker-8B
5. Save the key and copy the generated key
6. Send a rerank request like this:

curl -X POST "https://<proxy>/rerank" \
  -H "Authorization: Bearer <VIRTUAL_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "Qwen3-Reranker-8B",
    "query": "ping",
    "documents": [
      "say ping",
      "say pong",
      "hello world"
    ],
    "top_n": 2,
    "mock_testing_fallbacks": true
  }'

7. Observe that the request fails immediately instead of routing to the configured fallback

Relevant log output

{
  "error": {
    "message": "litellm.InternalServerError: This is a mock exception for model=Qwen3-Reranker-8B, to trigger a fallback. Fallbacks=None. Received Model Group=Qwen3-Reranker-8B\nAvailable Model Group Fallbacks=None",
    "type": null,
    "param": null,
    "code": "500"
  }
}

What part of LiteLLM is this about?

Proxy

What LiteLLM version are you on ?

v1.83.3

Twitter / LinkedIn details

No response