[Bug]: Google Gemini Flash 2.0 - Parameter "presence_penalty" not supported

[JFingerle]

What happened?

UnsupportedParamsError occurs when the "presence_penalty" parameter is used with Gemini Flash 2.0 using the Google API.

• gemini/gemini-2.0-flash-001

• gemini/gemini-2.0-flash

No error occurs via the openrouter API ( openrouter/google/gemini-2.0-flash-001)

Relevant log output

litellm.UnsupportedParamsError: gemini does not support parameters: ['presence_penalty'], for model=gemini-2.0-flash-001. To drop these, set `litellm.drop_params=True` or for proxy:

`litellm_settings:
 drop_params: true`

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.77.3.dynamic_rates

Twitter / LinkedIn details

No response

[JFingerle]

Workaround

In the litellm config set the setting allowed_openai_params in litellm_params of the model like this:

- model_name: gemini-2.0-flash-free
    litellm_params:
      model: gemini/gemini-2.0-flash-001
      api_key: secret

      # Workaroud for github bug #14753
      allowed_openai_params: ['frequency_penalty', 'logit_bias', 'logprobs', 'top_logprobs', 'max_tokens', 'max_completion_tokens', 'modalities', 'prediction', 'n', 'presence_penalty', 'seed', 'stop', 'stream', 'stream_options', 'temperature', 'top_p', 'tools', 'tool_choice', 'function_call', 'functions', 'max_retries', 'extra_headers', 'parallel_tool_calls', 'audio', 'web_search_options', 'safety_identifier', 'response_format']
    model_info:
      max_input_tokens: 1048576
      max_output_tokens: 8192
      context_length: 1048576
      input_cost_per_token: 0
      output_cost_per_token: 0
      rpm: 15       # requests per minute
      tpm: 1000000  # tokens per minute
      weight: 2     # smaller number is less preferred