diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py index 5228f656dba..f02333c34c8 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py @@ -13,14 +13,18 @@ from typing import ( List, Literal, Optional, + get_args, ) +from typing_extensions import assert_never + from litellm._logging import verbose_logger from litellm._uuid import uuid from litellm.types.llms.anthropic import ( AppliedEdit, CompactionBlock, ContextManagementResponse, + StreamingContentBlockDeltaType, UsageDelta, UsageIteration, ) @@ -30,6 +34,23 @@ if TYPE_CHECKING: from litellm.types.utils import ModelResponseStream +_STREAMING_DELTA_TYPES = frozenset(get_args(StreamingContentBlockDeltaType)) + + +def _delta_payload_field(delta_type: StreamingContentBlockDeltaType) -> str: + match delta_type: + case "text_delta": + return "text" + case "input_json_delta": + return "partial_json" + case "thinking_delta": + return "thinking" + case "signature_delta": + return "signature" + case _: + assert_never(delta_type) + + class _CombinedChunkSplitter: """ Splits a streaming chunk that carries BOTH response content and a @@ -837,6 +858,12 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): so the trigger chunk's delta must be re-queued or the first token of the new block (the first non-empty text/thinking delta, or bundled tool arguments) is silently dropped. + + Delta types outside ``StreamingContentBlockDeltaType`` — the closed + set the translate layer can produce — are treated as empty. The + per-type payload lookup is exhaustively matched against that set in + ``_delta_payload_field``, so extending the translate layer with a new + delta type fails type-checking here until it is handled. """ if processed_chunk.get("type") != "content_block_delta": return False @@ -844,15 +871,9 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): if not isinstance(delta, dict): return False delta_type = delta.get("type") - if delta_type == "text_delta": - return bool(delta.get("text")) - if delta_type == "input_json_delta": - return bool(delta.get("partial_json")) - if delta_type == "thinking_delta": - return bool(delta.get("thinking")) - if delta_type == "signature_delta": - return bool(delta.get("signature")) - return False + if delta_type not in _STREAMING_DELTA_TYPES: + return False + return bool(delta.get(_delta_payload_field(delta_type))) def _should_start_new_content_block(self, chunk: "ModelResponseStream") -> bool: """ diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py index 4c981dd36b3..cd75eed2e6e 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py @@ -104,6 +104,7 @@ from litellm.types.llms.anthropic import ( ContextManagementResponse, MessageBlockDelta, MessageDelta, + StreamingContentBlockDeltaType, UsageDelta, UsageIteration, ) @@ -1423,7 +1424,7 @@ class LiteLLMAnthropicMessagesAdapter: def _translate_streaming_openai_chunk_to_anthropic( self, choices: List[Union[OpenAIStreamingChoice, StreamingChoices]] ) -> Tuple[ - Literal["text_delta", "input_json_delta", "thinking_delta", "signature_delta"], + StreamingContentBlockDeltaType, Union[ ContentTextBlockDelta, ContentJsonBlockDelta, diff --git a/litellm/types/llms/anthropic.py b/litellm/types/llms/anthropic.py index aa9f4dccbd1..c24d072217a 100644 --- a/litellm/types/llms/anthropic.py +++ b/litellm/types/llms/anthropic.py @@ -439,6 +439,9 @@ class ContentThinkingSignatureBlockDelta(TypedDict): signature: str +StreamingContentBlockDeltaType = Literal["text_delta", "input_json_delta", "thinking_delta", "signature_delta"] + + class ContentBlockDelta(TypedDict): type: Literal["content_block_delta"] index: int