litellm.drop_params error when running the openapi server

[masaruduy]

I'm running the server normally with: python -m routellm.openai_server --routers mf --weak-model ollama_chat/codeqwen
and am getting this whenever I attempt a prompt:

File "C:\Users\nate\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\litellm\utils.py", line 3586, in get_optional_params
_check_valid_arg(supported_params=supported_params)
File "C:\Users\nate\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.11_qbz5n2kfra8p0\LocalCache\local-packages\Python311\site-packages\litellm\utils.py", line 3060, in _check_valid_arg
raise UnsupportedParamsError(
litellm.exceptions.UnsupportedParamsError: litellm.UnsupportedParamsError: ollama_chat does not support parameters: {'presence_penalty': 0.0}, for model=codeqwen. To drop these, set litellm.drop_params=True or for proxy:

litellm_settings: drop_params: true

I've tried modifying the yaml to no avail.
Please help!

[masaruduy]

oh and am trying to use it with continue.dev
{
"model": "router-mf-0.11593",
"title": "routellm",
"completionOptions": {},
"apiBase": "http://192.168.0.215:6060/v1",
"apiKey": "no_api_key",
"provider": "openai"
},

[smwaqas89]

Hi @masaruduy, I've submitted a fix for this in the linked PR.

Root cause: ChatCompletionRequest in openai_server.py defaults presence_penalty and frequency_penalty to 0.0 (not None), so model_dump(exclude_none=True) always includes them. When litellm forwards these to providers like Ollama that don't support them, it raises UnsupportedParamsError.

Fix: Set litellm.drop_params = True in controller.py, which is litellm's built-in mechanism for silently dropping unsupported parameters. This fixes the issue for all non-OpenAI providers (Ollama, Groq, local models, etc.).

This also addresses the same underlying issue from PR #32.