vllm localhost fails to accept json

the openai format provided has the following format for the body:
{
"model": "model_id",
"parameters": {
"max_new_tokens": 60,
...
},
"prompt":"text"
}

however the accepted format in VLLM does not have a parameters field. it should be of the format:
{
"model": "model_id",
"max_new_tokens": 60,
...
"prompt":"text"
}

When you try to set the request body and don't include the parameters as a field then it is automatically added.
This then throws a 400 request error.

No files, tests, or entry points are named. Start by tracing where localhost vLLM/OpenAI-compatible request bodies are assembled, then verify the request shape and its 400-response behavior; done means an existing parameters field is preserved and is not added automatically.

Written by the indexing model from the issue text.