mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-25 05:54:53 +00:00
427 lines
17 KiB
Python
427 lines
17 KiB
Python
import inspect
|
|
from collections.abc import Awaitable, Callable, Mapping
|
|
from typing import Final, cast # noqa: TID251 # narrows legacy callable signatures for inspect
|
|
|
|
import pytest
|
|
|
|
import litellm
|
|
from litellm.llms.anthropic.pass_through.messages import handler as python_messages
|
|
from litellm.messages import dispatch
|
|
from litellm.messages.dispatch import (
|
|
_ADISPATCH, # pyright: ignore[reportPrivateUsage] # tests configured dispatch
|
|
_DISPATCH, # pyright: ignore[reportPrivateUsage] # tests configured dispatch
|
|
)
|
|
from litellm.proxy import proxy_server
|
|
from litellm.rust_bridge.bindings import NativeBinding
|
|
from litellm.rust_bridge.catalog import Feature, Route, RouteRule, Rules, Target
|
|
from litellm.rust_bridge.configuration import Rollout
|
|
from litellm.rust_bridge.messages.entrypoints import NATIVE_AMESSAGES, NATIVE_MESSAGES, NativeAmessages, NativeMessages
|
|
from litellm.rust_bridge.public_call import NativeCall
|
|
from litellm.types.llms.anthropic_messages.anthropic_response import AnthropicMessagesResponse
|
|
|
|
MESSAGES: Final = [{"role": "user", "content": "hi"}]
|
|
PYTHON_RULES: Final[Rules] = ()
|
|
RUST_RULES: Final[Rules] = (RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED),)
|
|
|
|
|
|
def messages_binding(native: NativeMessages | None) -> NativeBinding[NativeMessages]:
|
|
binding: Final[NativeBinding[NativeMessages]] = NativeBinding("anthropic_messages_handler", validate=lambda _: None)
|
|
binding.override(native)
|
|
return binding
|
|
|
|
|
|
def amessages_binding(native: NativeAmessages | None) -> NativeBinding[NativeAmessages]:
|
|
binding: Final[NativeBinding[NativeAmessages]] = NativeBinding("anthropic_messages", validate=lambda _: None)
|
|
binding.override(native)
|
|
return binding
|
|
|
|
|
|
def response(model: str = "claude-sonnet-4-5") -> AnthropicMessagesResponse:
|
|
return AnthropicMessagesResponse(id="msg_test", type="message", role="assistant", model=model, content=[])
|
|
|
|
|
|
def test_public_signature_is_the_legacy_signature() -> None:
|
|
public_messages: Final = cast(Callable[..., object], litellm.anthropic_messages_handler)
|
|
legacy_messages: Final = cast(Callable[..., object], python_messages.anthropic_messages_handler)
|
|
public_amessages: Final = cast(Callable[..., object], litellm.anthropic_messages)
|
|
legacy_amessages: Final = cast(Callable[..., object], python_messages.anthropic_messages)
|
|
assert inspect.signature(public_messages) == inspect.signature(legacy_messages)
|
|
assert inspect.signature(public_amessages) == inspect.signature(legacy_amessages)
|
|
|
|
|
|
def test_python_route_forwards_original_call_shape() -> None:
|
|
metadata: Final = {"user_id": "u"}
|
|
args: Final[tuple[object, ...]] = (16, MESSAGES, "claude-sonnet-4-5")
|
|
kwargs: Final[Mapping[str, object]] = {"temperature": 0.1, "litellm_metadata": metadata}
|
|
captured: Final[list[tuple[tuple[object, ...], Mapping[str, object]]]] = []
|
|
expected: Final = response()
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse: # kwargs-ok: records call shape
|
|
captured.append((call_args, call_kwargs))
|
|
return expected
|
|
|
|
def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
pytest.fail("Python-only dispatch must not call native")
|
|
|
|
result: Final = _DISPATCH.run(
|
|
args,
|
|
kwargs,
|
|
python=python,
|
|
binding=messages_binding(native),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=PYTHON_RULES,
|
|
)
|
|
assert result is expected
|
|
call_args, call_kwargs = captured[0]
|
|
assert call_args == args
|
|
assert call_args[1] is MESSAGES
|
|
assert call_kwargs == kwargs
|
|
assert call_kwargs["litellm_metadata"] is metadata
|
|
assert kwargs == {"temperature": 0.1, "litellm_metadata": metadata}
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_async_python_route_forwards_original_call_shape() -> None:
|
|
metadata: Final = {"user_id": "u"}
|
|
args: Final[tuple[object, ...]] = (16, MESSAGES, "claude-sonnet-4-5")
|
|
kwargs: Final[Mapping[str, object]] = {"temperature": 0.1, "litellm_metadata": metadata}
|
|
captured: Final[list[tuple[tuple[object, ...], Mapping[str, object]]]] = []
|
|
expected: Final = response()
|
|
|
|
async def python(
|
|
*call_args: object,
|
|
**call_kwargs: object, # kwargs-ok: records call shape
|
|
) -> AnthropicMessagesResponse:
|
|
captured.append((call_args, call_kwargs))
|
|
return expected
|
|
|
|
async def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
pytest.fail("Python-only dispatch must not call native")
|
|
|
|
result: Final = await _ADISPATCH.arun(
|
|
args,
|
|
kwargs,
|
|
python=python,
|
|
binding=amessages_binding(native),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=PYTHON_RULES,
|
|
)
|
|
assert result is expected
|
|
call_args, call_kwargs = captured[0]
|
|
assert call_args == args
|
|
assert call_args[1] is MESSAGES
|
|
assert call_kwargs == kwargs
|
|
assert call_kwargs["litellm_metadata"] is metadata
|
|
assert kwargs == {"temperature": 0.1, "litellm_metadata": metadata}
|
|
|
|
|
|
def test_native_receives_normalized_request_and_original_call_shape() -> None:
|
|
metadata: Final = {"user_id": "u"}
|
|
args: Final[tuple[object, ...]] = (16, MESSAGES, "anthropic/claude-sonnet-4-5")
|
|
kwargs: Final[Mapping[str, object]] = {
|
|
"stream": True,
|
|
"api_key": "sk-test",
|
|
"api_base": "https://example.invalid",
|
|
"custom_llm_provider": "anthropic",
|
|
"litellm_metadata": metadata,
|
|
}
|
|
captured: Final[list[tuple[NativeCall, tuple[object, ...], Mapping[str, object]]]] = []
|
|
expected: Final = response("anthropic/claude-sonnet-4-5")
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse: # kwargs-ok: rejected fallback
|
|
pytest.fail("Required Rust dispatch must not call Python")
|
|
|
|
def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
args: Final = request.args
|
|
kwargs: Final = request.kwargs
|
|
captured.append((request, args, kwargs))
|
|
return expected
|
|
|
|
result: Final = _DISPATCH.run(
|
|
args,
|
|
kwargs,
|
|
python=python,
|
|
binding=messages_binding(native),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=RUST_RULES,
|
|
)
|
|
assert result is expected
|
|
request, call_args, call_kwargs = captured[0]
|
|
assert request.resolved["model"] == "anthropic/claude-sonnet-4-5"
|
|
assert request.resolved["messages"] is MESSAGES
|
|
assert request.resolved["max_tokens"] == 16
|
|
assert request.resolved["stream"] is True
|
|
assert request.resolved["api_key"] == "sk-test"
|
|
assert request.resolved["api_base"] == "https://example.invalid"
|
|
assert request.resolved["custom_llm_provider"] == "anthropic"
|
|
assert request.kwargs == kwargs
|
|
assert request.kwargs["litellm_metadata"] is metadata
|
|
assert call_args == args
|
|
assert call_args[1] is MESSAGES
|
|
assert call_kwargs == kwargs
|
|
assert call_kwargs["litellm_metadata"] is metadata
|
|
|
|
|
|
@pytest.mark.parametrize("stream, expected_timeout", ((True, 0.5), (False, 30.0)))
|
|
def test_native_streams_are_bounded_by_stream_timeout(stream: bool, expected_timeout: float) -> None:
|
|
captured: Final[list[NativeCall]] = []
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse: # kwargs-ok: rejected fallback
|
|
pytest.fail("Required Rust dispatch must not call Python")
|
|
|
|
def native(request: NativeCall) -> AnthropicMessagesResponse:
|
|
captured.append(request)
|
|
return response()
|
|
|
|
_ = _DISPATCH.run(
|
|
(16, MESSAGES, "anthropic/claude-sonnet-4-5"),
|
|
{"stream": stream, "timeout": 30.0, "stream_timeout": 0.5},
|
|
python=python,
|
|
binding=messages_binding(native),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=RUST_RULES,
|
|
)
|
|
assert captured[0].resolved["timeout"] == expected_timeout
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("model", "kwargs", "passes_through"),
|
|
(
|
|
("vertex_ai/claude-sonnet-4-5@20250929", {}, True),
|
|
("claude-sonnet-4-5@20250929", {"custom_llm_provider": "vertex_ai"}, True),
|
|
("deepseek/deepseek-v4-flash", {}, True),
|
|
("openai/gpt-5", {"model_info": {"supported_endpoints": ["/v1/messages"]}}, True),
|
|
("vertex_ai/gemini-2.5-pro", {}, False),
|
|
("gemini-2.5-pro", {"custom_llm_provider": "vertex_ai"}, False),
|
|
("openai/gpt-5", {}, False),
|
|
("groq/llama-3.3-70b-versatile", {}, False),
|
|
("unknown-provider/unknown-model", {}, False),
|
|
),
|
|
)
|
|
def test_only_requests_that_pass_through_to_a_messages_api_reach_a_messages_target_rule(
|
|
model: str, kwargs: Mapping[str, object], passes_through: bool
|
|
) -> None:
|
|
python_result: Final = response(model)
|
|
native_result: Final = response(model)
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse: # kwargs-ok: fallback
|
|
return python_result
|
|
|
|
result: Final = _DISPATCH.run(
|
|
(16, MESSAGES, model),
|
|
kwargs,
|
|
python=python,
|
|
binding=messages_binding(lambda request: native_result),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=(RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED, targets=frozenset({Target.MESSAGES})),),
|
|
)
|
|
assert result is (native_result if passes_through else python_result)
|
|
|
|
|
|
DANGEROUS_TOOL_USE: Final = [{"type": "dangerous_tool_use", "classifier_context": {"permission_mode": "auto"}}]
|
|
CLASSIFIER_SETTINGS: Final = {"safeguards_classifier_model": "gpt-oss-120b"}
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("model", "kwargs", "general_settings", "serves_python"),
|
|
(
|
|
("deepseek/deepseek-v4-flash", {"safeguards": DANGEROUS_TOOL_USE}, CLASSIFIER_SETTINGS, True),
|
|
(
|
|
"deepseek-v4-flash",
|
|
{"safeguards": DANGEROUS_TOOL_USE, "custom_llm_provider": "deepseek"},
|
|
CLASSIFIER_SETTINGS,
|
|
True,
|
|
),
|
|
("deepseek/deepseek-v4-flash", {"safeguards": DANGEROUS_TOOL_USE}, {}, False),
|
|
("deepseek/deepseek-v4-flash", {}, CLASSIFIER_SETTINGS, False),
|
|
("deepseek/deepseek-v4-flash", {"safeguards": [{"type": "other"}]}, CLASSIFIER_SETTINGS, False),
|
|
("anthropic/claude-sonnet-4-5", {"safeguards": DANGEROUS_TOOL_USE}, CLASSIFIER_SETTINGS, False),
|
|
),
|
|
)
|
|
def test_proxy_classified_safeguards_reach_the_feature_rule(
|
|
monkeypatch: pytest.MonkeyPatch,
|
|
model: str,
|
|
kwargs: Mapping[str, object],
|
|
general_settings: Mapping[str, object],
|
|
serves_python: bool,
|
|
) -> None:
|
|
monkeypatch.setattr(proxy_server, "general_settings", general_settings)
|
|
python_result: Final = response(model)
|
|
native_result: Final = response(model)
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse: # kwargs-ok: fallback
|
|
return python_result
|
|
|
|
result: Final = _DISPATCH.run(
|
|
(16, MESSAGES, model),
|
|
kwargs,
|
|
python=python,
|
|
binding=messages_binding(lambda request: native_result),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=(
|
|
RouteRule(Route.MESSAGES, Rollout.PYTHON_ONLY, features=frozenset({Feature.SAFEGUARDS_CLASSIFIER})),
|
|
RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED),
|
|
),
|
|
)
|
|
assert result is (python_result if serves_python else native_result)
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("args", "kwargs"),
|
|
(
|
|
((16, MESSAGES, "claude-sonnet-4-5"), {"model": "duplicate"}),
|
|
((), {}),
|
|
),
|
|
)
|
|
def test_binding_errors_delegate_to_python(args: tuple[object, ...], kwargs: Mapping[str, object]) -> None:
|
|
captured: Final[list[tuple[tuple[object, ...], Mapping[str, object]]]] = []
|
|
expected: Final = response()
|
|
|
|
def python(*call_args: object, **call_kwargs: object) -> AnthropicMessagesResponse:
|
|
captured.append((call_args, call_kwargs))
|
|
return expected
|
|
|
|
def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
pytest.fail("Binding failures must be delegated to Python")
|
|
|
|
result: Final = _DISPATCH.run(
|
|
args,
|
|
kwargs,
|
|
python=python,
|
|
binding=messages_binding(native),
|
|
native=lambda hook, request, call_args, call_kwargs: hook(request),
|
|
rules=RUST_RULES,
|
|
)
|
|
assert result is expected
|
|
assert captured == [(args, kwargs)]
|
|
|
|
|
|
def test_anthropic_create_routes_through_dispatch() -> None:
|
|
captured: Final[list[NativeCall]] = []
|
|
expected: Final = response()
|
|
|
|
def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
captured.append(request)
|
|
return expected
|
|
|
|
NATIVE_MESSAGES.override(native)
|
|
public_create: Final = cast(Callable[..., AnthropicMessagesResponse], litellm.anthropic.create)
|
|
try:
|
|
result: Final = public_create(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5")
|
|
finally:
|
|
NATIVE_MESSAGES.reset()
|
|
assert result is expected
|
|
assert [request.resolved["model"] for request in captured] == ["claude-sonnet-4-5"]
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_anthropic_acreate_routes_through_dispatch() -> None:
|
|
captured: Final[list[NativeCall]] = []
|
|
expected: Final = response()
|
|
|
|
async def native(
|
|
request: NativeCall,
|
|
) -> AnthropicMessagesResponse:
|
|
captured.append(request)
|
|
return expected
|
|
|
|
NATIVE_AMESSAGES.override(native)
|
|
public_acreate: Final = cast(Callable[..., Awaitable[AnthropicMessagesResponse]], litellm.anthropic.acreate)
|
|
try:
|
|
result: Final = await public_acreate(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5")
|
|
finally:
|
|
NATIVE_AMESSAGES.reset()
|
|
assert result is expected
|
|
assert [request.resolved["model"] for request in captured] == ["claude-sonnet-4-5"]
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_truthy_is_async_hands_off_to_the_async_entry_like_true() -> None:
|
|
captured: Final[list[NativeCall]] = []
|
|
expected: Final = response()
|
|
|
|
async def native(request: NativeCall) -> AnthropicMessagesResponse:
|
|
captured.append(request)
|
|
return expected
|
|
|
|
def sync_native(request: NativeCall) -> AnthropicMessagesResponse:
|
|
pytest.fail("a truthy is_async must hand off to the async entry, as the Python handler does")
|
|
|
|
NATIVE_AMESSAGES.override(native)
|
|
NATIVE_MESSAGES.override(sync_native)
|
|
public_create: Final = cast(Callable[..., object], litellm.anthropic.create)
|
|
try:
|
|
pending: Final = public_create(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5", is_async=1)
|
|
assert inspect.isawaitable(pending)
|
|
assert await pending is expected
|
|
reference: Final = public_create(max_tokens=16, messages=MESSAGES, model="claude-sonnet-4-5", is_async=True)
|
|
assert inspect.isawaitable(reference)
|
|
assert await reference is expected
|
|
finally:
|
|
NATIVE_AMESSAGES.reset()
|
|
NATIVE_MESSAGES.reset()
|
|
assert [dict(request.resolved) for request in captured] == [dict(captured[1].resolved)] * 2
|
|
|
|
|
|
def test_sync_messages_request_projects_public_arguments() -> None:
|
|
rules: Final[Rules] = (RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED),)
|
|
expected: Final = AnthropicMessagesResponse(model="claude-test")
|
|
|
|
def native(request: NativeCall) -> AnthropicMessagesResponse:
|
|
assert request.resolved["model"] == "claude-test"
|
|
assert request.resolved["messages"] == MESSAGES
|
|
assert request.resolved["max_tokens"] == 10
|
|
assert request.resolved["custom_llm_provider"] == "anthropic"
|
|
return expected
|
|
|
|
binding: Final[NativeBinding[NativeMessages]] = NativeBinding("messages", validate=lambda _: None)
|
|
binding.override(native)
|
|
response: Final = dispatch._DISPATCH.run( # pyright: ignore[reportPrivateUsage] # test an explicit route decision
|
|
(),
|
|
{
|
|
"model": "claude-test",
|
|
"messages": MESSAGES,
|
|
"max_tokens": 10,
|
|
"custom_llm_provider": "anthropic",
|
|
},
|
|
python=lambda *args, **kwargs: pytest.fail("required native route must handle this call"),
|
|
binding=binding,
|
|
native=lambda hook, request, args, kwargs: hook(request),
|
|
rules=rules,
|
|
)
|
|
|
|
assert response is expected
|
|
|
|
|
|
def test_messages_binding_error_delegates_unchanged_to_python() -> None:
|
|
rules: Final[Rules] = (RouteRule(Route.MESSAGES, Rollout.RUST_REQUIRED),)
|
|
expected: Final = AnthropicMessagesResponse(model="claude-test")
|
|
|
|
def python(*args: object, **kwargs: object) -> AnthropicMessagesResponse:
|
|
return expected
|
|
|
|
def native(request: NativeCall) -> AnthropicMessagesResponse:
|
|
pytest.fail("a call without max_tokens cannot project a request and must stay on Python")
|
|
|
|
binding: Final[NativeBinding[NativeMessages]] = NativeBinding("messages", validate=lambda _: None)
|
|
binding.override(native)
|
|
response: Final = dispatch._DISPATCH.run( # pyright: ignore[reportPrivateUsage] # test an explicit route decision
|
|
(),
|
|
{"model": "claude-test", "messages": MESSAGES, "custom_llm_provider": "anthropic"},
|
|
python=python,
|
|
binding=binding,
|
|
native=lambda hook, request, args, kwargs: hook(request),
|
|
rules=rules,
|
|
)
|
|
|
|
assert response is expected
|