OpenAI-compat: OpenAiInferenceEngine base_url /v1 with logprobs off

[cursor]

Working config (OpenAiInferenceEngine + base_url /v1)

Unitxt OpenAiInferenceEngine posts Chat Completions. base_url is the /v1 root (or OPENAI_API_URL). Default logprobs=True is not supported on every Chat Completions host. CrossProviderInferenceEngine only lists configured providers. AzureOpenAIInferenceEngine is Azure.

A host whose public catalog is GET https://api.pzero.studio/v1/models (no key) fits:

import os
from unitxt.inference import OpenAiInferenceEngine

os.environ["OPENAI_API_KEY"] = "<Bearer key for that host>"

engine = OpenAiInferenceEngine(
    model_name="deepseek-v4-flash",
    base_url="https://api.pzero.studio/v1",
    logprobs=False,
)

model_name is a catalog id. Do not paste /v1/chat/completions as base_url.

Not asking for a first-party provider. OpenAiInferenceEngine plus base_url plus logprobs=False is enough.