[Bug]: drop_params works on individual models but not on global litellm_settings

[colindonovan-8451]

What happened?

We have run into issues when using the OpenAI library to query non-OpenAI embeddings models through LiteLLM. We have previously instituted a fix for this problem by adding drop_params=true to all effected embeddings models, but recently we tried to move that setting to the global litellm_settings block outside of our model_list. Unfortunately, when we moved this setting up a level, we began to see the error that we had seen before again.

Relevant log output

"litellm.UnsupportedParamsError: vertex_ai does not support parameters: {'encoding_format': 'base64'}, for model=text-embedding-005. To drop these, set `litellm.drop_params=True` or for proxy:\n\n`litellm_settings:\n drop_params: true`\

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.55.9

Twitter / LinkedIn details

No response

[krrishdholakia]

Hey @colindonovan-8451 just ran this and it works fine for me.

How did you set it on your config?

[ishaan-jaff]

does this still occur @colindonovan-8451 ?

[ishaan-jaff]

can you share how you set this on your config ?

[colindonovan-8451]

Sorry for the delay - missed the initial response.

We were seeing this behavior on 1.55.9 when we set drop_params: true under the litellm_settings block at the top level of our config yaml. We ended up adding drop_params: true to all individual models to get around this error.

[ishaan-jaff]

Additional notes:

• setting drop_params globally did not work for vertex text embeddings
• setting model specific did work
• User is using OpenAI python SDK

[krrishdholakia]

closing as unable to repro.

[krrishdholakia]

curl used to test:

curl -L -X POST 'http://0.0.0.0:4000/embeddings' \
-H 'Authorization: Bearer sk-1234' \
-H 'Content-Type: application/json' \
-d '{"input": ["hello world"], "model": "text-embedding-005", "encoding_format": "base64"}'

config.yaml

model_list:
  - model_name: text-embedding-005
    litellm_params:
      model: vertex_ai/text-embedding-005

litellm_settings:
  cache: true
  drop_params: true