diff --git a/litellm/llms/bedrock/chat/native_output_fallback.py b/litellm/llms/bedrock/chat/native_output_fallback.py index 1b8ce5fcda4..6892cd37beb 100644 --- a/litellm/llms/bedrock/chat/native_output_fallback.py +++ b/litellm/llms/bedrock/chat/native_output_fallback.py @@ -1,11 +1,14 @@ from collections.abc import Callable, Coroutine +from functools import partial from typing import Final, TypeAlias from litellm.llms.base_llm.chat.transformation import BaseLLMException from litellm.llms.bedrock.common_utils import ( NativeOutputFallback, is_bedrock_native_output_rejection, + is_bedrock_strict_tools_rejection, warn_bedrock_native_output_rejected, + warn_bedrock_strict_tools_dropped, ) from litellm.types.utils import ModelResponse from litellm.utils import CustomStreamWrapper @@ -14,12 +17,13 @@ NativeOutputResponse: TypeAlias = ModelResponse | CustomStreamWrapper NativeOutputAttemptResult: TypeAlias = NativeOutputResponse | Coroutine[object, object, NativeOutputResponse] NativeOutputAttempt: TypeAlias = Callable[[NativeOutputFallback | None], NativeOutputAttemptResult] NativeOutputFallbackFactory: TypeAlias = Callable[[], NativeOutputFallback | None] +_RetryFor: TypeAlias = Callable[[BaseLLMException], NativeOutputFallback | None] -def _fallback_for( - error: BaseLLMException, fallback: NativeOutputFallbackFactory, model: str +def _native_output_retry( + error: BaseLLMException, fallback: NativeOutputFallbackFactory | None, model: str ) -> NativeOutputFallback | None: - if not is_bedrock_native_output_rejection(error.status_code, error.message): + if fallback is None or not is_bedrock_native_output_rejection(error.status_code, error.message): return None retry: Final = fallback() if retry is not None: @@ -27,16 +31,38 @@ def _fallback_for( return retry +def _strict_tools_retry( + error: BaseLLMException, fallback: NativeOutputFallbackFactory | None, model: str +) -> NativeOutputFallback | None: + if fallback is None or not is_bedrock_strict_tools_rejection(error.status_code, error.message): + return None + retry: Final = fallback() + if retry is not None: + warn_bedrock_strict_tools_dropped(model) + return retry + + +def _fallback_for( + error: BaseLLMException, + fallback: NativeOutputFallbackFactory | None, + strict_tools_fallback: NativeOutputFallbackFactory | None, + model: str, +) -> NativeOutputFallback | None: + native_output_retry: Final = _native_output_retry(error, fallback, model) + if native_output_retry is not None: + return native_output_retry + return _strict_tools_retry(error, strict_tools_fallback, model) + + async def _await_with_native_output_fallback( first_attempt: Coroutine[object, object, NativeOutputResponse], attempt: NativeOutputAttempt, - fallback: NativeOutputFallbackFactory, - model: str, + retry_for: _RetryFor, ) -> NativeOutputResponse: try: return await first_attempt except BaseLLMException as error: - retry: Final = _fallback_for(error, fallback, model) + retry: Final = retry_for(error) if retry is None: raise second_attempt: Final = attempt(retry) @@ -47,16 +73,20 @@ def call_with_native_output_fallback( attempt: NativeOutputAttempt, fallback: NativeOutputFallbackFactory | None, model: str, + strict_tools_fallback: NativeOutputFallbackFactory | None = None, ) -> NativeOutputAttemptResult: - if fallback is None: + if fallback is None and strict_tools_fallback is None: return attempt(None) + retry_for: Final = partial( + _fallback_for, fallback=fallback, strict_tools_fallback=strict_tools_fallback, model=model + ) try: first_attempt: Final = attempt(None) except BaseLLMException as error: - retry: Final = _fallback_for(error, fallback, model) + retry: Final = retry_for(error) if retry is None: raise return attempt(retry) if isinstance(first_attempt, Coroutine): - return _await_with_native_output_fallback(first_attempt, attempt, fallback, model) + return _await_with_native_output_fallback(first_attempt, attempt, retry_for) return first_attempt diff --git a/litellm/llms/bedrock/common_utils.py b/litellm/llms/bedrock/common_utils.py index 0cf3e895ad2..c508514373c 100644 --- a/litellm/llms/bedrock/common_utils.py +++ b/litellm/llms/bedrock/common_utils.py @@ -359,6 +359,44 @@ class NativeOutputFallback: litellm_params: Mapping[str, object] +_BEDROCK_STRICT_TOOLS_REJECTION: Final = re.compile( + r"strict tools|grammar|compilation|tools\.\d+\.custom(?:\.strict)?: " +) +_TOOLS_ADAPTER: Final = TypeAdapter(tuple[Mapping[str, object], ...]) +_TOOL_FUNCTION_ADAPTER: Final = TypeAdapter(Mapping[str, object]) + + +def is_bedrock_strict_tools_rejection(status_code: int, error_text: str) -> bool: + return status_code == 400 and _BEDROCK_STRICT_TOOLS_REJECTION.search(error_text) is not None + + +def warn_bedrock_strict_tools_dropped(model: str) -> None: + verbose_logger.warning( + "Bedrock rejected the strict tool schemas for model=%s, so litellm sent the request again without `strict` " + "because drop_params is set.", + model, + ) + + +def _without_strict(tool: Mapping[str, object]) -> Mapping[str, object]: + if "function" not in tool: + return tool + function: Final = _TOOL_FUNCTION_ADAPTER.validate_python(tool["function"]) + return {**tool, "function": {key: value for key, value in function.items() if key != "strict"}} + + +def bedrock_strict_tools_fallback( + optional_params: Mapping[str, object], litellm_params: Mapping[str, object] +) -> NativeOutputFallback | None: + tools: Final = _TOOLS_ADAPTER.validate_python(optional_params.get("tools", ())) + tools_without_strict: Final = tuple(_without_strict(tool) for tool in tools) + if tools_without_strict == tools: + return None + return NativeOutputFallback( + optional_params={**optional_params, "tools": list(tools_without_strict)}, litellm_params=litellm_params + ) + + def requests_output_config_format(optional_params: Mapping[str, object]) -> bool: match optional_params: case {"output_config": {"format": object()}}: diff --git a/litellm/main.py b/litellm/main.py index d647cab24e3..f6c2bec40f6 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -121,6 +121,7 @@ from litellm.llms.bedrock.common_utils import ( BedrockModelInfo, NativeOutputFallback, bedrock_route_for_request, + bedrock_strict_tools_fallback, deployment_model_info, without_bedrock_native_output_opt_in, without_bedrock_route_prefix, @@ -4353,6 +4354,11 @@ def _complete_bedrock(ctx: CompletionDispatchContext) -> _CompletionDispatchResu else None ), converse_model, + strict_tools_fallback=( + partial(bedrock_strict_tools_fallback, converse_params, dict(litellm_params)) + if litellm.drop_params is True or normalize_drop_params(litellm_params.get("drop_params")) is True + else None + ), ) elif bedrock_route == "converse_like": model = model.replace("converse_like/", "") diff --git a/tests/unit/llms/bedrock/chat/test_native_output_fallback.py b/tests/unit/llms/bedrock/chat/test_native_output_fallback.py index 301fdaeb245..4e2bd9069b8 100644 --- a/tests/unit/llms/bedrock/chat/test_native_output_fallback.py +++ b/tests/unit/llms/bedrock/chat/test_native_output_fallback.py @@ -121,3 +121,86 @@ def test_a_rejection_logs_the_retry_and_advises_opting_out_only_when_native_outp assert _MODEL in caplog.text assert ("supports_bedrock_native_structured_output: false" in caplog.text) is advises_opting_out + + +_STRICT_FALLBACK: Final = NativeOutputFallback(optional_params={"tools": "without_strict"}, litellm_params={}) +# Bedrock's 400s for strict tools, seen live on us.anthropic.claude-sonnet-4-6 on 2026-10-20 and quoted in issue #34388, +# each answered 200 once the request dropped strict +_STRICT_TOOLS_REJECTIONS: Final = tuple( + BedrockError(status_code=400, message=f'{{"message":"The model returned the following errors: {text}"}}') + for text in ( + "Too many strict tools (21). The maximum number of strict tools supported is 20.", + "Schemas contains too many optional parameters (27), which would make grammar compilation inefficient.", + "Schemas contains too many parameters with union types (18 parameters with type arrays or anyOf). " + "This causes exponential compilation cost.", + "The compiled grammar is too large, which would cause performance issues.", + "tools.0.custom: Invalid schema: Schema type is missing for schema: {'description': 'tag'}", + "tools.3.custom: For 'integer' type, property 'minimum' is not supported", + "Compiled grammar size (410.4MB) exceeds maximum allowed size (300MB)", + "tools.1.custom.strict: Extra inputs are not permitted", + ) +) +_TOO_MANY_STRICT_TOOLS: Final = _STRICT_TOOLS_REJECTIONS[0] + + +def _call_with_strict_tools_fallback( + bedrock: ScriptedBedrock, fallback: NativeOutputFallback | None, strict_tools_fallback: NativeOutputFallback | None +) -> object: + return _resolve( + call_with_native_output_fallback( + bedrock, + None if fallback is None else lambda: fallback, + _MODEL, + strict_tools_fallback=None if strict_tools_fallback is None else lambda: strict_tools_fallback, + ) + ) + + +@_MODES +@pytest.mark.parametrize("rejection", _STRICT_TOOLS_REJECTIONS, ids=lambda error: error.message[60:100]) +def test_a_strict_tools_rejection_is_sent_again_once_with_the_strict_tools_fallback( + asynchronous: bool, rejection: BedrockError +) -> None: + bedrock: Final = ScriptedBedrock(errors=(rejection, None), asynchronous=asynchronous) + + response: Final = _call_with_strict_tools_fallback(bedrock, None, _STRICT_FALLBACK) + + assert isinstance(response, ModelResponse) and response.id == "answer-2" + assert bedrock.sent == (None, _STRICT_FALLBACK) + + +@_MODES +@pytest.mark.parametrize( + ("error", "strict_tools_fallback"), + [ + (_OTHER_400, _STRICT_FALLBACK), + (BedrockError(status_code=500, message=_TOO_MANY_STRICT_TOOLS.message), _STRICT_FALLBACK), + (_TOO_MANY_STRICT_TOOLS, None), + ], + ids=["other_400", "server_error", "no_strict_tools_fallback"], +) +def test_no_second_request_without_both_a_strict_tools_rejection_and_a_strict_tools_fallback( + asynchronous: bool, error: BedrockError, strict_tools_fallback: NativeOutputFallback | None +) -> None: + bedrock: Final = ScriptedBedrock(errors=(error,), asynchronous=asynchronous) + + with pytest.raises(BedrockError) as raised: + _call_with_strict_tools_fallback(bedrock, None, strict_tools_fallback) + + assert raised.value is error + assert bedrock.sent == (None,) + + +@pytest.mark.parametrize( + ("error", "retried_with"), + [(_REJECTION, _FALLBACK), (_TOO_MANY_STRICT_TOOLS, _STRICT_FALLBACK)], + ids=["native_output_rejection", "strict_tools_rejection"], +) +def test_each_rejection_is_sent_again_with_its_own_fallback_when_both_are_set( + error: BedrockError, retried_with: NativeOutputFallback +) -> None: + bedrock: Final = ScriptedBedrock(errors=(error, None), asynchronous=False) + + _call_with_strict_tools_fallback(bedrock, _FALLBACK, _STRICT_FALLBACK) + + assert bedrock.sent == (None, retried_with) diff --git a/tests/unit/llms/bedrock/test_bedrock_common_utils.py b/tests/unit/llms/bedrock/test_bedrock_common_utils.py index 7bde1805e18..4bc450b3ad9 100644 --- a/tests/unit/llms/bedrock/test_bedrock_common_utils.py +++ b/tests/unit/llms/bedrock/test_bedrock_common_utils.py @@ -6,7 +6,9 @@ from typing import Final from litellm.llms.bedrock.common_utils import( AmazonBedrockGlobalConfig, BedrockModelInfo, + NativeOutputFallback, bedrock_native_output_is_opt_in, + bedrock_strict_tools_fallback, deployment_model_info, extract_model_name_from_bedrock_arn, get_bedrock_base_model, @@ -2150,3 +2152,41 @@ def test_bedrock_response_stream_shape_is_unavailable_when_botocore_describes_th assert shape is None (logged,) = warning.call_args_list assert "only-in-the-description" not in logged.args[0] % logged.args[1:] + + +def test_bedrock_strict_tools_fallback_drops_only_strict_and_keeps_everything_else(): + strict_tool: Final = { + "type": "function", + "cache_control": {"type": "ephemeral"}, + "function": {"name": "lookup", "description": "Look up", "strict": True, "parameters": {"type": "object"}}, + } + plain_tool: Final = {"type": "function", "function": {"name": "plain", "parameters": {"type": "object"}}} + anthropic_tool: Final = {"name": "native", "input_schema": {"type": "object"}} + litellm_params: Final = {"drop_params": True} + + fallback: Final = bedrock_strict_tools_fallback( + {"tools": [strict_tool, plain_tool, anthropic_tool], "maxTokens": 16, "stream": True}, litellm_params + ) + + assert fallback == NativeOutputFallback( + optional_params={ + "tools": [ + {**strict_tool, "function": {"name": "lookup", "description": "Look up", "parameters": {"type": "object"}}}, + plain_tool, + anthropic_tool, + ], + "maxTokens": 16, + "stream": True, + }, + litellm_params=litellm_params, + ) + assert strict_tool["function"]["strict"] is True + + +@pytest.mark.parametrize( + "optional_params", + [{"maxTokens": 16}, {"tools": [{"type": "function", "function": {"name": "plain", "parameters": {}}}]}], + ids=["no_tools", "no_strict_tool"], +) +def test_bedrock_strict_tools_fallback_is_none_without_a_strict_tool(optional_params): + assert bedrock_strict_tools_fallback(optional_params, {}) is None