What happened?
I have the meta-llama/Meta-Llama-3-8b-Instruct model exposed using vLLM with the OpenAI API interface. I'm attempting to evaluate my system using Phoenix, but I need to utilize LiteLLM to interact with my vLLM model. This is the link to Phoenic docs (https://docs.arize.com/phoenix/api/evaluation-models#litellmmodel)
However, I'm encountering an issue where the litellm.completion function adds an extra_body parameter to the request, which causes an error in vLLM.
My code snippet that throws error
os.environ["OPENAI_API_KEY"] = "api_key"
eval_model = LiteLLMModel(
model="openai/meta-llama/Meta-Llama-3-8b-Instruct",
model_kwargs=dict(
api_base="http://llama3-8b:8000/v1/chat/completions",
),
)
Function Call in LiteLLM:
litellm.completion(
model='openai/meta-llama/Meta-Llama-3-8b-Instruct',
messages=[{'content': 'Tell me a joke', 'role': 'user'}],
temperature=0.0,
max_tokens=256,
top_p=1,
num_retries=0,
request_timeout=60,
api_base='http://llama3-8b:8000/v1/chat/completions'
)
cURL Request Sent from LiteLLM:
curl -X POST \
http://llama3-8b:8000/v1/chat/completions/ \
-d '{'model': 'meta-llama/Meta-Llama-3-8b-Instruct', 'messages': [{'content': 'Tell me a joke?', 'role': 'user'}], 'temperature': 0.0, 'top_p': 1, 'max_tokens': 256, 'extra_body': {}}'
Error Response from vLLM:
{
"object": "error",
"message": "[{'type': 'extra_forbidden', 'loc': ('body', 'extra_body'), 'msg': 'Extra inputs are not permitted', 'input': {}}]",
"type": "BadRequestError",
"param": null,
"code": 400
}
Request:
I need a way to configure or modify LiteLLM so that it does not include the extra_body parameter in the request, as this parameter is not accepted by vLLM and causes a BadRequestError.
Thank you for your assistance!
Relevant log output
No response
Twitter / LinkedIn details
No response
What happened?
I have the
meta-llama/Meta-Llama-3-8b-Instructmodel exposed using vLLM with the OpenAI API interface. I'm attempting to evaluate my system using Phoenix, but I need to utilize LiteLLM to interact with my vLLM model. This is the link to Phoenic docs (https://docs.arize.com/phoenix/api/evaluation-models#litellmmodel)However, I'm encountering an issue where the
litellm.completionfunction adds anextra_bodyparameter to the request, which causes an error in vLLM.My code snippet that throws error
Function Call in LiteLLM:
cURL Request Sent from LiteLLM:
Error Response from vLLM:
{ "object": "error", "message": "[{'type': 'extra_forbidden', 'loc': ('body', 'extra_body'), 'msg': 'Extra inputs are not permitted', 'input': {}}]", "type": "BadRequestError", "param": null, "code": 400 }Request:
I need a way to configure or modify LiteLLM so that it does not include the
extra_bodyparameter in the request, as this parameter is not accepted by vLLM and causes aBadRequestError.Thank you for your assistance!
Relevant log output
No response
Twitter / LinkedIn details
No response