The fastest, litest AI Gateway. Rust core with Python SDK. Call 100+ LLM APIs in OpenAI (or native) format with cost tracking, guardrails, load balancing, and logging [Bedrock, Azure, OpenAI, Anthropic, OpenAI, VertexAI, vLLM, Nvidia NIM]
Python ai-gatewayanthropicazure-openaibedrockgatewaylangchainlitellmllmllm-gatewayllmopsmcp-gatewayopenaiopenai-proxyrustrust-aivertex-ai
import json
from typing import Final, Literal
import httpx
import pytest
import respx
import litellm
from litellm.types.utils import LiteLLMFineTuningJob
API_BASE: Final = "https://fine-tuning.test"
BASE_MODEL: Final = "gpt-4.1-nano-2025-04-14"
PROVIDER_JOB: Final = {
"id": "ftjob-abc123",
"object": "fine_tuning.job",
"created_at": 1700000000,
"error": None,
"fine_tuned_model": None,
"finished_at": None,
"hyperparameters": {"n_epochs": 3},
"model": BASE_MODEL,
"organization_id": "org-test",
"result_files": [],
"seed": 42,
"status": "validating_files",
"trained_tokens": None,
"training_file": "file-abc123",
"validation_file": None,
}
@pytest.mark.asyncio
@respx.mock
@pytest.mark.parametrize(
("custom_llm_provider", "route"),
[("openai", f"{API_BASE}/v1/fine_tuning/jobs"), ("azure", f"{API_BASE}/openai/fine_tuning/jobs")],
)
@pytest.mark.parametrize("prefixed", [True, False], ids=["provider-prefixed", "bare"])
async def test_create_sends_the_model_without_the_provider_prefix(
monkeypatch: pytest.MonkeyPatch, custom_llm_provider: str, route: str, prefixed: bool
) -> None:
sent: Final = respx.post(url__startswith=route).mock(return_value=httpx.Response(200, json=PROVIDER_JOB))
async with httpx.AsyncClient() as session:
monkeypatch.setattr(litellm, "aclient_session", session)
await litellm.acreate_fine_tuning_job(
model=f"{custom_llm_provider}/{BASE_MODEL}" if prefixed else BASE_MODEL,
training_file="file-abc123",
custom_llm_provider=custom_llm_provider,
api_key="sk-test",
api_base=f"{API_BASE}/v1" if custom_llm_provider == "openai" else API_BASE,
**({"api_version": "2024-10-21"} if custom_llm_provider == "azure" else {}),
)
assert json.loads(sent.calls.last.request.content)["model"] == BASE_MODEL
@pytest.mark.asyncio
@respx.mock
@pytest.mark.parametrize(
("custom_llm_provider", "api_base", "route", "sent_api_version"),
[
("openai", f"{API_BASE}/v1", f"{API_BASE}/v1/fine_tuning/jobs", None),
("azure", API_BASE, f"{API_BASE}/openai/fine_tuning/jobs", "2024-10-21"),
],
)
@pytest.mark.parametrize("is_async", [True, False], ids=["async", "sync"])
async def test_create_with_api_version_reaches_the_provider_and_only_azure_gets_it(
monkeypatch: pytest.MonkeyPatch,
custom_llm_provider: Literal["openai", "azure"],
api_base: str,
route: str,
sent_api_version: str | None,
is_async: bool,
) -> None:
sent: Final = respx.post(url__startswith=route).mock(return_value=httpx.Response(200, json=PROVIDER_JOB))
request: Final = {
"model": BASE_MODEL,
"training_file": "file-abc123",
"custom_llm_provider": custom_llm_provider,
"api_key": "sk-test",
"api_base": api_base,
"api_version": "2024-10-21",
}
async with httpx.AsyncClient() as session:
monkeypatch.setattr(litellm, "aclient_session", session)
job: Final = (
await litellm.acreate_fine_tuning_job(**request) if is_async else litellm.create_fine_tuning_job(**request)
)
assert isinstance(job, LiteLLMFineTuningJob)
assert job.id == PROVIDER_JOB["id"]
assert sent.calls.last.request.url.params.get("api-version") == sent_api_version