tiennm99/litellm
1
0
Fork 0
mirror of https://github.com/tiennm99/litellm.git synced 2026-08-05 18:24:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(responses): rename bridge opt-in to use_chat_completions_api

Browse Source 
- Add use_chat_completions_api (keep use_responses_api_bridge as deprecated alias)
- Support openai/chat_completions/<model> model prefix for the same behavior
- Forward use_chat_completions_api in file_search emulation inner calls
- Update response_api.md and extend unit tests

Made-with: Cursor
This commit is contained in:
Sameer Kankute
2026-04-13 18:03:27 +05:30
parent 6b7629ec04
commit 0a8bf4ec9e
4 changed files with 124 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/my-website/docs/response_api.md
+25 -4
View File
@@ -1509,11 +1509,15 @@ curl http://localhost:4000/v1/responses \
If you're using an **OpenAI-compatible third-party provider** (e.g. llama.cpp, vLLM, LM Studio) via `openai/` prefix with a custom `api_base`, LiteLLM will normally forward `/responses` requests directly to that endpoint. If the provider only supports `/chat/completions`, the request will fail.
Set `use_responses_api_bridge: true` to force the `/responses` → `/chat/completions` bridge for these models.
Use any of these to force the `/responses` → `/chat/completions` bridge:
1. **`use_chat_completions_api: true`** (recommended) — makes it explicit that LiteLLM will call the provider’s chat-completions API.
2. **`openai/chat_completions/<model_name>`** — same pattern as `responses/` on chat completions: the model id encodes the routing choice.
3. **`use_responses_api_bridge: true`** — deprecated alias for `use_chat_completions_api` (kept for backward compatibility).
#### Python SDK Usage
```python showLineNumbers title="Force bridge for custom openai/ endpoint"
```python showLineNumbers title="Force bridge for custom openai/ endpoint (flag)"
import litellm
response = litellm.responses(
@@ -1521,7 +1525,22 @@ response = litellm.responses(
input="Hello!",
api_base="http://localhost:8080",
api_key="fake-key",
use_responses_api_bridge=True,
use_chat_completions_api=True,
)
print(response)
```
Or encode it in the model id:
```python showLineNumbers title="Force bridge via openai/chat_completions/ model prefix"
import litellm
response = litellm.responses(
model="openai/chat_completions/my-custom-model",
input="Hello!",
api_base="http://localhost:8080",
api_key="fake-key",
)
print(response)
@@ -1538,9 +1557,11 @@ model_list:
model: openai/my-custom-model
api_base: http://localhost:8080/v1
api_key: fake-key
use_responses_api_bridge: true
use_chat_completions_api: true
```
Alternatively set `model: openai/chat_completions/my-custom-model` instead of the flag.
**Start Proxy:**
```bash showLineNumbers title="Start LiteLLM Proxy"
litellm/responses/main.py
+40 -4
View File
@@ -643,6 +643,29 @@ def _apply_prompt_management_to_responses_call(
return input, model, custom_llm_provider
# Opt-in via model id (mirrors the `responses/` prefix pattern on chat completions).
_OPENAI_CHAT_COMPLETIONS_RESPONSES_MODEL_PREFIX = "openai/chat_completions/"
def _normalize_openai_chat_completions_responses_model(model: str) -> tuple[str, bool]:
"""
Strip `openai/chat_completions/<name>` → `openai/<name>` and return True when the
prefix was applied (same effect as use_chat_completions_api=True).
"""
if not model.startswith(_OPENAI_CHAT_COMPLETIONS_RESPONSES_MODEL_PREFIX):
return model, False
remainder = model[len(_OPENAI_CHAT_COMPLETIONS_RESPONSES_MODEL_PREFIX) :]
if not remainder:
return model, False
return f"openai/{remainder}", True
def _pop_use_chat_completions_api_kw(kwargs: Dict[str, Any]) -> bool:
"""Pop bridge flags; True if either requests the chat-completions path."""
use_cc = kwargs.pop("use_chat_completions_api", None)
return bool(use_cc)
def _resolve_model_provider_for_responses(
model: str,
custom_llm_provider: Optional[str],
@@ -754,7 +777,7 @@ def responses(
litellm_logging_obj: LiteLLMLoggingObj = kwargs.get("litellm_logging_obj") # type: ignore
litellm_call_id: Optional[str] = kwargs.get("litellm_call_id", None)
_is_async = kwargs.pop("aresponses", False) is True
use_responses_api_bridge = kwargs.pop("use_responses_api_bridge", None)
use_chat_completions_api = _pop_use_chat_completions_api_kw(kwargs)
# Convert text_format to text parameter if provided
text = ResponsesAPIRequestUtils.convert_text_format_to_text_param(
@@ -777,6 +800,15 @@ def responses(
mock_response=litellm_params.mock_response
)
_stripped_model, _from_chat_completions_prefix = (
_normalize_openai_chat_completions_responses_model(model)
)
model = _stripped_model
local_vars["model"] = model
use_chat_completions_api = (
use_chat_completions_api or _from_chat_completions_prefix
)
model, custom_llm_provider = _resolve_model_provider_for_responses(
model=model,
custom_llm_provider=custom_llm_provider,
@@ -872,7 +904,7 @@ def responses(
if _has_file_search_tool(tools) and (
responses_api_provider_config is None
or use_responses_api_bridge is True
or use_chat_completions_api is True
or not responses_api_provider_config.supports_native_file_search()
):
from litellm.responses.file_search.emulated_handler import (
@@ -907,7 +939,11 @@ def responses(
"extra_body": extra_body,
"timeout": timeout,
"custom_llm_provider": custom_llm_provider,
**({"use_responses_api_bridge": True} if use_responses_api_bridge else {}),
**(
{"use_chat_completions_api": True}
if use_chat_completions_api
else {}
),
**{k: v for k, v in kwargs.items() if k not in _internal_skip},
}
if _is_async:
@@ -922,7 +958,7 @@ def responses(
**emulated_kwargs,
)
if responses_api_provider_config is None or use_responses_api_bridge is True:
if responses_api_provider_config is None or use_chat_completions_api is True:
return litellm_completion_transformation_handler.response_api_handler(
model=model,
input=input,
litellm/types/router.py
+3 -3
View File
@@ -199,7 +199,7 @@ class GenericLiteLLMParams(CredentialLiteLLMParams, CustomPricingLiteLLMParams):
budget_duration: Optional[str] = None
use_in_pass_through: Optional[bool] = False
use_litellm_proxy: Optional[bool] = False
use_responses_api_bridge: Optional[bool] = None
use_chat_completions_api: Optional[bool] = None
model_config = ConfigDict(extra="allow", arbitrary_types_allowed=True)
merge_reasoning_content_in_choices: Optional[bool] = False
model_info: Optional[Dict] = None
@@ -319,8 +319,8 @@ class LiteLLMParamsTypedDict(TypedDict, total=False):
configurable_clientside_auth_params: CONFIGURABLE_CLIENTSIDE_AUTH_PARAMS # for allowing api base switching on finetuned models
## DROP PARAMS ##
drop_params: Optional[bool]
## RESPONSES API BRIDGE ##
use_responses_api_bridge: Optional[bool]
## RESPONSES API → CHAT COMPLETIONS BRIDGE ##
use_chat_completions_api: Optional[bool]
## UNIFIED PROJECT/REGION ##
region_name: Optional[str]
## VERTEX AI ##
tests/test_litellm/responses/test_responses_api_bridge_flag.py
+56 -9
View File
@@ -1,6 +1,7 @@
"""
Tests for the `use_responses_api_bridge` flag that allows openai/ models
with custom api_base to opt-in to the /responses → /chat/completions bridge.
Tests for forcing the /responses → /chat/completions bridge for `openai/` models
(via `use_chat_completions_api`, deprecated `use_responses_api_bridge`, or the
`openai/chat_completions/<model>` model id).
Includes file_search emulation: the flag must be forwarded on inner aresponses
calls so routed requests do not hit a custom api_base /v1/responses endpoint.
@@ -19,7 +20,7 @@ from litellm.types.llms.openai import ResponseAPIUsage, ResponsesAPIResponse
class TestUseResponsesApiBridgeFlag:
"""Test that use_responses_api_bridge forces the chat completions bridge."""
"""Test that bridge opt-in forces the chat completions path."""
@patch(
"litellm.responses.main.litellm_completion_transformation_handler.response_api_handler"
@@ -28,8 +29,7 @@ class TestUseResponsesApiBridgeFlag:
"litellm.responses.main.ProviderConfigManager.get_provider_responses_api_config"
)
def test_bridge_used_when_flag_is_true(self, mock_get_config, mock_bridge_handler):
"""When use_responses_api_bridge=True, the bridge handler should be called
even though the provider (openai) has native responses API support."""
"""When use_responses_api_bridge=True (deprecated alias), the bridge runs."""
# Setup: provider config returns a non-None config (native support exists)
mock_get_config.return_value = litellm.OpenAIResponsesAPIConfig()
@@ -44,6 +44,51 @@ class TestUseResponsesApiBridgeFlag:
mock_bridge_handler.assert_called_once()
@patch(
"litellm.responses.main.litellm_completion_transformation_handler.response_api_handler"
)
@patch(
"litellm.responses.main.ProviderConfigManager.get_provider_responses_api_config"
)
def test_bridge_used_when_use_chat_completions_api_true(
self, mock_get_config, mock_bridge_handler
):
"""When use_chat_completions_api=True, the bridge handler should be called."""
mock_get_config.return_value = litellm.OpenAIResponsesAPIConfig()
mock_bridge_handler.return_value = MagicMock()
litellm.responses(
model="openai/my-custom-model",
input="Hello",
use_chat_completions_api=True,
litellm_logging_obj=MagicMock(),
)
mock_bridge_handler.assert_called_once()
@patch(
"litellm.responses.main.litellm_completion_transformation_handler.response_api_handler"
)
@patch(
"litellm.responses.main.ProviderConfigManager.get_provider_responses_api_config"
)
def test_bridge_used_when_model_uses_chat_completions_prefix(
self, mock_get_config, mock_bridge_handler
):
"""`openai/chat_completions/<name>` normalizes to `openai/<name>` and uses the bridge."""
mock_get_config.return_value = litellm.OpenAIResponsesAPIConfig()
mock_bridge_handler.return_value = MagicMock()
litellm.responses(
model="openai/chat_completions/my-custom-model",
input="Hello",
litellm_logging_obj=MagicMock(),
)
mock_bridge_handler.assert_called_once()
# Model string is provider-normalized after resolution; prefix only forces the bridge.
assert mock_bridge_handler.call_args.kwargs["model"].endswith("my-custom-model")
@patch("litellm.responses.main.base_llm_http_handler.response_api_handler")
@patch(
"litellm.responses.main.ProviderConfigManager.get_provider_responses_api_config"
@@ -84,9 +129,10 @@ class TestUseResponsesApiBridgeFlag:
)
call_kwargs = mock_bridge_handler.call_args
# The flag should not appear in the kwargs passed to the bridge handler
# Bridge flags should not appear in the kwargs passed to the bridge handler
all_kwargs = call_kwargs.kwargs if call_kwargs.kwargs else {}
assert "use_responses_api_bridge" not in all_kwargs
assert "use_chat_completions_api" not in all_kwargs
@patch(
"litellm.responses.main.litellm_completion_transformation_handler.response_api_handler"
@@ -145,12 +191,12 @@ class TestUseResponsesApiBridgeFlag:
litellm_logging_obj=MagicMock(),
)
# Verify _call_aresponses was called with use_responses_api_bridge=True
# Verify _call_aresponses was called with use_chat_completions_api=True
mock_call_aresponses.assert_called_once()
call_kwargs = mock_call_aresponses.call_args.kwargs
assert (
call_kwargs.get("use_responses_api_bridge") is True
), "use_responses_api_bridge flag should be forwarded to inner aresponses call"
call_kwargs.get("use_chat_completions_api") is True
), "use_chat_completions_api should be forwarded to inner aresponses call"
@patch(
"litellm.responses.main.litellm_completion_transformation_handler.response_api_handler"
@@ -223,6 +269,7 @@ class TestUseResponsesApiBridgeFlag:
for call in mock_bridge_handler.call_args_list:
all_kwargs = call.kwargs if call.kwargs else {}
assert "use_responses_api_bridge" not in all_kwargs
assert "use_chat_completions_api" not in all_kwargs
assert result is not None
assert result.id is not None