Merge pull request #24687 from Sameerlite/litellm_litellm_azure-finetuning-fixes

feat(fine-tuning): fix Azure OpenAI fine-tuning job creation
This commit is contained in:
yuneng-jiang 2026-03-27 09:58:12 -07:00 committed by GitHub
commit f3fe6d1c0a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 331 additions and 29 deletions

View file

@ -1,10 +1,15 @@
from typing import Optional, Union
from typing import Any, Coroutine, Dict, Optional, Union, cast
import httpx
from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
from litellm._logging import verbose_logger
from litellm.llms.azure.common_utils import BaseAzureLLM
from litellm.llms.openai.fine_tuning.handler import OpenAIFineTuningAPI
from litellm.llms.openai.fine_tuning.handler import (
OpenAIFineTuningAPI,
_litellm_fine_tuning_job_from_response,
)
from litellm.types.utils import LiteLLMFineTuningJob
class AzureOpenAIFineTuningAPI(OpenAIFineTuningAPI, BaseAzureLLM):
@ -12,6 +17,194 @@ class AzureOpenAIFineTuningAPI(OpenAIFineTuningAPI, BaseAzureLLM):
AzureOpenAI methods to support fine tuning, inherits from OpenAIFineTuningAPI.
"""
@staticmethod
def _ensure_training_type(create_fine_tuning_job_data: Dict[str, Any]) -> None:
"""
Azure requires trainingType in extra_body. Default to 1 (supervised) if omitted.
"""
extra_body = create_fine_tuning_job_data.get("extra_body") or {}
if not isinstance(extra_body, dict):
extra_body = {}
if extra_body.get("trainingType") is None:
extra_body["trainingType"] = 1
create_fine_tuning_job_data["extra_body"] = extra_body
verbose_logger.debug(
"Azure fine-tuning: defaulting trainingType=1 (supervised)"
)
async def acreate_fine_tuning_job(
self,
create_fine_tuning_job_data: dict,
openai_client: Union[AsyncOpenAI, AsyncAzureOpenAI],
) -> LiteLLMFineTuningJob:
response = await openai_client.fine_tuning.jobs.create(
**create_fine_tuning_job_data
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
async def acancel_fine_tuning_job(
self,
fine_tuning_job_id: str,
openai_client: Union[AsyncOpenAI, AsyncAzureOpenAI],
) -> LiteLLMFineTuningJob:
response = await openai_client.fine_tuning.jobs.cancel(
fine_tuning_job_id=fine_tuning_job_id
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
async def aretrieve_fine_tuning_job(
self,
fine_tuning_job_id: str,
openai_client: Union[AsyncOpenAI, AsyncAzureOpenAI],
) -> LiteLLMFineTuningJob:
response = await openai_client.fine_tuning.jobs.retrieve(
fine_tuning_job_id=fine_tuning_job_id
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
def create_fine_tuning_job(
self,
_is_async: bool,
create_fine_tuning_job_data: dict,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
organization: Optional[str],
client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = None,
) -> Union[LiteLLMFineTuningJob, Coroutine[Any, Any, LiteLLMFineTuningJob]]:
self._ensure_training_type(create_fine_tuning_job_data)
openai_client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = self.get_openai_client(
api_key=api_key,
api_base=api_base,
timeout=timeout,
max_retries=max_retries,
organization=organization,
client=client,
_is_async=_is_async,
api_version=api_version,
)
if openai_client is None:
raise ValueError(
"Azure OpenAI client is not initialized. Make sure api_key is passed or AZURE_API_KEY is set in the environment."
)
if _is_async is True:
if not isinstance(openai_client, (AsyncOpenAI, AsyncAzureOpenAI)):
raise ValueError(
"OpenAI client is not an instance of AsyncOpenAI. Make sure you passed an AsyncOpenAI client."
)
return self.acreate_fine_tuning_job(
create_fine_tuning_job_data=create_fine_tuning_job_data,
openai_client=openai_client,
)
verbose_logger.debug(
"creating fine tuning job, args= %s", create_fine_tuning_job_data
)
response = cast(OpenAI, openai_client).fine_tuning.jobs.create(
**create_fine_tuning_job_data
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
def cancel_fine_tuning_job(
self,
_is_async: bool,
fine_tuning_job_id: str,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
organization: Optional[str],
client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = None,
) -> Union[LiteLLMFineTuningJob, Coroutine[Any, Any, LiteLLMFineTuningJob]]:
openai_client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = self.get_openai_client(
api_key=api_key,
api_base=api_base,
timeout=timeout,
max_retries=max_retries,
organization=organization,
client=client,
_is_async=_is_async,
api_version=api_version,
)
if openai_client is None:
raise ValueError(
"Azure OpenAI client is not initialized. Make sure api_key is passed or AZURE_API_KEY is set in the environment."
)
if _is_async is True:
if not isinstance(openai_client, (AsyncOpenAI, AsyncAzureOpenAI)):
raise ValueError(
"OpenAI client is not an instance of AsyncOpenAI. Make sure you passed an AsyncOpenAI client."
)
return self.acancel_fine_tuning_job(
fine_tuning_job_id=fine_tuning_job_id,
openai_client=openai_client,
)
response = cast(OpenAI, openai_client).fine_tuning.jobs.cancel(
fine_tuning_job_id=fine_tuning_job_id
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
def retrieve_fine_tuning_job(
self,
_is_async: bool,
fine_tuning_job_id: str,
api_key: Optional[str],
api_base: Optional[str],
api_version: Optional[str],
timeout: Union[float, httpx.Timeout],
max_retries: Optional[int],
organization: Optional[str],
client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = None,
) -> Union[LiteLLMFineTuningJob, Coroutine[Any, Any, LiteLLMFineTuningJob]]:
openai_client: Optional[
Union[OpenAI, AsyncOpenAI, AzureOpenAI, AsyncAzureOpenAI]
] = self.get_openai_client(
api_key=api_key,
api_base=api_base,
timeout=timeout,
max_retries=max_retries,
organization=organization,
client=client,
_is_async=_is_async,
api_version=api_version,
)
if openai_client is None:
raise ValueError(
"Azure OpenAI client is not initialized. Make sure api_key is passed or AZURE_API_KEY is set in the environment."
)
if _is_async is True:
if not isinstance(openai_client, (AsyncOpenAI, AsyncAzureOpenAI)):
raise ValueError(
"OpenAI client is not an instance of AsyncOpenAI. Make sure you passed an AsyncOpenAI client."
)
return self.aretrieve_fine_tuning_job(
fine_tuning_job_id=fine_tuning_job_id,
openai_client=openai_client,
)
response = cast(OpenAI, openai_client).fine_tuning.jobs.retrieve(
fine_tuning_job_id=fine_tuning_job_id
)
return _litellm_fine_tuning_job_from_response(response, is_azure=True)
def get_openai_client(
self,
api_key: Optional[str],

View file

@ -1,4 +1,4 @@
from typing import Any, Coroutine, Optional, Union, cast
from typing import Any, Coroutine, Dict, Optional, Union, cast
import httpx
from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
@ -6,6 +6,55 @@ from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
from litellm._logging import verbose_logger
from litellm.types.utils import LiteLLMFineTuningJob
_AZURE_STATUS_MAP = {
"pending": "queued",
"notRunning": "queued",
"running": "running",
"succeeded": "succeeded",
"failed": "failed",
"canceled": "cancelled",
"canceling": "cancelled",
}
# Note: Azure's "canceling" (in-progress) is mapped to "cancelled" (terminal)
# because LiteLLMFineTuningJob schema has no intermediate cancellation state.
def _normalize_fine_tuning_job_dict(
data: Dict[str, Any], is_azure: bool = False
) -> Dict[str, Any]:
"""
Normalize Azure OpenAI FineTuningJob response to match OpenAI schema.
Azure differences:
- organization_id: null ""
- result_files: null []
- status: mapped via _AZURE_STATUS_MAP
"""
if not is_azure:
return data
normalized = data.copy()
if normalized.get("organization_id") is None:
normalized["organization_id"] = ""
if normalized.get("result_files") is None:
normalized["result_files"] = []
status = normalized.get("status")
if status in _AZURE_STATUS_MAP:
normalized["status"] = _AZURE_STATUS_MAP[status]
return normalized
def _litellm_fine_tuning_job_from_response(
response: Any, is_azure: bool = False
) -> LiteLLMFineTuningJob:
return LiteLLMFineTuningJob(
**_normalize_fine_tuning_job_dict(response.model_dump(), is_azure=is_azure)
)
class OpenAIFineTuningAPI:
"""
@ -60,7 +109,7 @@ class OpenAIFineTuningAPI:
**create_fine_tuning_job_data
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)
def create_fine_tuning_job(
self,
@ -108,7 +157,7 @@ class OpenAIFineTuningAPI:
response = cast(OpenAI, openai_client).fine_tuning.jobs.create(
**create_fine_tuning_job_data
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)
async def acancel_fine_tuning_job(
self,
@ -118,7 +167,7 @@ class OpenAIFineTuningAPI:
response = await openai_client.fine_tuning.jobs.cancel(
fine_tuning_job_id=fine_tuning_job_id
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)
def cancel_fine_tuning_job(
self,
@ -164,7 +213,7 @@ class OpenAIFineTuningAPI:
response = cast(OpenAI, openai_client).fine_tuning.jobs.cancel(
fine_tuning_job_id=fine_tuning_job_id
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)
async def alist_fine_tuning_jobs(
self,
@ -229,7 +278,7 @@ class OpenAIFineTuningAPI:
response = await openai_client.fine_tuning.jobs.retrieve(
fine_tuning_job_id=fine_tuning_job_id
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)
def retrieve_fine_tuning_job(
self,
@ -275,4 +324,4 @@ class OpenAIFineTuningAPI:
response = cast(OpenAI, openai_client).fine_tuning.jobs.retrieve(
fine_tuning_job_id=fine_tuning_job_id
)
return LiteLLMFineTuningJob(**response.model_dump())
return _litellm_fine_tuning_job_from_response(response)

View file

@ -315,11 +315,11 @@ class OpenAIFileObject(BaseModel):
`fine-tune`, `fine-tune-results`, `vision`, and `user_data`.
"""
status: Optional[Literal["uploaded", "processed", "error"]] = None
status: Optional[Literal["uploaded", "processed", "error", "pending"]] = None
"""Deprecated.
The current status of the file, which can be either `uploaded`, `processed`, or
`error`.
The current status of the file, which can be either `uploaded`, `processed`,
`error`, or `pending` (Azure may return `pending` immediately after upload).
"""
expires_at: Optional[int] = None

View file

@ -180,6 +180,21 @@ async def test_azure_create_fine_tune_jobs_async():
pass
def test_azure_trainingtype_defaults_to_one():
"""
Azure requires trainingType in extra_body. When omitted, AzureOpenAIFineTuningAPI defaults it to 1.
"""
from litellm.llms.azure.fine_tuning.handler import AzureOpenAIFineTuningAPI
handler = AzureOpenAIFineTuningAPI()
create_data = {"model": "gpt-4o-mini", "training_file": "file-test"}
handler._ensure_training_type(create_data)
assert "extra_body" in create_data
assert create_data["extra_body"]["trainingType"] == 1
@pytest.mark.asyncio()
async def test_create_vertex_fine_tune_jobs_mocked():
load_vertex_ai_credentials()
@ -601,11 +616,10 @@ async def test_mock_openai_retrieve_fine_tune_job():
@pytest.mark.asyncio
async def test_mock_azure_create_fine_tune_job_with_azure_specific_params():
"""Test that Azure-specific parameters are passed through extra_body"""
from openai import AsyncAzureOpenAI
from openai.types.fine_tuning.fine_tuning_job import FineTuningJob
from openai.types.fine_tuning.fine_tuning_job import Hyperparameters as OAIHyperparameters
from litellm.types.utils import LiteLLMFineTuningJob
mock_response = FineTuningJob(
mock_response = LiteLLMFineTuningJob(
id="ft-azure-123",
model="gpt-4.1-mini-2025-04-14",
created_at=1677610602,
@ -619,8 +633,11 @@ async def test_mock_azure_create_fine_tune_job_with_azure_specific_params():
result_files=[],
)
async def mock_async_create(*args, **kwargs):
return mock_response
with patch("litellm.llms.azure.fine_tuning.handler.AzureOpenAIFineTuningAPI.create_fine_tuning_job") as mock_create:
mock_create.return_value = mock_response
mock_create.return_value = mock_async_create()
response = await litellm.acreate_fine_tuning_job(
model="gpt-4.1-mini-2025-04-14",

View file

@ -219,11 +219,13 @@ class TestAssistantMessageImageUrlContent:
# convert to list to consume it — this must not raise ValidationError.
content_blocks = list(raw_content) if raw_content is not None else []
assert len(content_blocks) == 2, (
f"Expected 2 content blocks (text + image_url), got {len(content_blocks)}: {content_blocks}"
)
assert (
len(content_blocks) == 2
), f"Expected 2 content blocks (text + image_url), got {len(content_blocks)}: {content_blocks}"
types = [b.get("type") for b in content_blocks if isinstance(b, dict)]
assert "image_url" in types, f"image_url block was silently dropped; blocks: {content_blocks}"
assert (
"image_url" in types
), f"image_url block was silently dropped; blocks: {content_blocks}"
def test_assistant_message_image_url_preserved_in_all_message_values(self):
"""
@ -255,14 +257,16 @@ class TestAssistantMessageImageUrlContent:
assert assistant is not None, "Assistant message missing after serialisation"
content = assistant.get("content", [])
assert isinstance(content, list), f"content should be a list, got {type(content)}"
assert len(content) == 2, (
f"Expected 2 content blocks (text + image_url), got {len(content)}: {content}"
)
assert isinstance(
content, list
), f"content should be a list, got {type(content)}"
assert (
len(content) == 2
), f"Expected 2 content blocks (text + image_url), got {len(content)}: {content}"
types = [b.get("type") for b in content if isinstance(b, dict)]
assert "image_url" in types, (
f"image_url block was silently dropped during AllMessageValues serialisation; blocks: {content}"
)
assert (
"image_url" in types
), f"image_url block was silently dropped during AllMessageValues serialisation; blocks: {content}"
class TestResponsesAPIReasoningNullFields:
@ -379,10 +383,14 @@ class TestResponsesAPIReasoningNullFields:
)
dumped = response.model_dump()
reasoning = [
o for o in dumped["output"] if isinstance(o, dict) and o.get("type") == "reasoning"
o
for o in dumped["output"]
if isinstance(o, dict) and o.get("type") == "reasoning"
][0]
message = [
o for o in dumped["output"] if isinstance(o, dict) and o.get("type") == "message"
o
for o in dumped["output"]
if isinstance(o, dict) and o.get("type") == "message"
][0]
assert "status" not in reasoning
assert "content" not in reasoning
@ -410,3 +418,38 @@ class TestResponsesAPIReasoningNullFields:
assert dumped["error"] is None
assert "instructions" in dumped
assert dumped["instructions"] is None
def test_normalize_fine_tuning_job_dict_maps_azure_pending():
from litellm.llms.openai.fine_tuning.handler import _normalize_fine_tuning_job_dict
out = _normalize_fine_tuning_job_dict(
{"organization_id": None, "result_files": None, "status": "pending"},
is_azure=True,
)
assert out["organization_id"] == ""
assert out["result_files"] == []
assert out["status"] == "queued"
def test_normalize_fine_tuning_job_dict_openai_unchanged():
from litellm.llms.openai.fine_tuning.handler import _normalize_fine_tuning_job_dict
data = {"organization_id": None, "result_files": None, "status": "pending"}
out = _normalize_fine_tuning_job_dict(data, is_azure=False)
assert out is data
def test_openai_file_object_accepts_pending_status():
from litellm.types.llms.openai import OpenAIFileObject
file_obj = OpenAIFileObject(
id="file-123",
bytes=1024,
created_at=1677610602,
filename="train.jsonl",
object="file",
purpose="fine-tune",
status="pending",
)
assert file_obj.status == "pending"