# LiteLLM proxy: exposes every local model over an OpenAI API on :4000,
# with automatic fallbacks. The pre-router (router.py) picks the model;
# LiteLLM handles delivery, retries, and logging.
model_list:
  - model_name: qwen3:8b
    litellm_params: { model: ollama_chat/qwen3:8b, api_base: http://127.0.0.1:11434 }
  - model_name: qwen3:14b
    litellm_params: { model: ollama_chat/qwen3:14b, api_base: http://127.0.0.1:11434 }
  - model_name: glm-4.7-flash
    litellm_params: { model: ollama_chat/glm-4.7-flash, api_base: http://127.0.0.1:11434 }
  - model_name: qwen3-coder:30b
    litellm_params: { model: ollama_chat/qwen3-coder:30b, api_base: http://127.0.0.1:11434 }
  - model_name: qwen3.6:35b-a3b
    litellm_params: { model: ollama_chat/qwen3.6:35b-a3b, api_base: http://127.0.0.1:11434 }
  - model_name: gpt-oss:20b
    litellm_params: { model: ollama_chat/gpt-oss:20b, api_base: http://127.0.0.1:11434 }
  - model_name: qwen3-vl:8b
    litellm_params: { model: ollama_chat/qwen3-vl:8b, api_base: http://127.0.0.1:11434 }
  - model_name: qwen3-vl:30b-a3b-instruct
    litellm_params: { model: ollama_chat/qwen3-vl:30b-a3b-instruct, api_base: http://127.0.0.1:11434 }
  # catch-all: any other model name (incl. :latest tags picked by `ollama launch`) -> ollama
  - model_name: "*"
    litellm_params: { model: "ollama_chat/*", api_base: http://127.0.0.1:11434 }

litellm_settings:
  drop_params: true          # tolerate params a given model doesn't support

router_settings:
  fallbacks:
    - { "glm-4.7-flash": ["qwen3-coder:30b"] }
    - { "qwen3-coder:30b": ["glm-4.7-flash"] }
    - { "qwen3.6:35b-a3b": ["gpt-oss:20b"] }
    - { "qwen3-vl:30b-a3b-instruct": ["qwen3-vl:8b"] }
    - { "qwen3:14b": ["qwen3:8b"] }