LiteLLM fails to connect to OpenAI-compatible backend (DMR) — conflicting behavior between model prefix and provider detection

Description

When configuring Bytebot’s LiteLLM proxy to communicate with a Docker Model Runner (DMR) providing an OpenAI-compatible API, the proxy fails to register a healthy deployment or returns NotFoundError / AuthenticationError depending on configuration.

Despite DMR responding correctly to requests, LiteLLM cannot successfully bridge requests through litellm-config.yaml.

my litellm-config.yaml : model_list:

• model_name: local-qwen3 litellm_params: model: openai/ai/qwen3-vl:latest api_base: http://model-runner.docker.internal/engines/qwen3-vl api_key: "local-dev-key" litellm_settings: drop_params: true

with this probe : docker logs -f bytebot-bytebot-llm-proxy-1

As soon as connect to bytebot at localhost:9992 I get this: LiteLLM: Proxy initialized with Config, Set models: local-qwen3 INFO: 172.18.0.5:56186 - "GET /model/info HTTP/1.1" 200 OK

I see local-qwen3 from the drop-down list. But as soon as i request a task in the interface like "open firefox" then I get a task error: 13:03:23 - LiteLLM Proxy:ERROR: common_request_processing.py:699 - litellm.proxy.proxy_server._handle_llm_api_exception(): Exception occured - litellm.NotFoundError: NotFoundError: OpenAIException - not found. Received Model Group=openai/ai/qwen3-vl:latest Available Model Group Fallbacks=None

The error means the LiteLLM proxy can reach the OpenAI-compatible API but the model name it’s asking for doesn’t exist on the DMR side due to the openai/ai/qwen3-vl:latest.

Am I the only one having this issue by trying to interface liteLLM with Docker Model Runner ?

Keep in mind, I am a noob in Ai topics, so helped by chatGPT to conclude on this :)

Many thanks for your answers.