This commit is contained in:
jesco 2026-10-29 03:38:15 +00:00 • committed by GitHub
commit 454e605fcd
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 78 additions and 14 deletions

View file

@ -19,7 +19,6 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import (
from litellm.llms.ollama.common_utils import OllamaResponseIdNormalizer
from litellm.types.llms.databricks import DatabricksTool
from litellm.types.llms.openai import (
ChatCompletionReasoningItem,
ChatCompletionThinkingBlock,
ImageURLListItem,
OpenAIModerationResponse,
@ -158,18 +157,14 @@ def _clear_later_replay_slice_metadata(choice: StreamingChoices) -> None:
del choice.enhancements
def _first_replay_delta(
message: Mapping[str, object], reasoning_items: Sequence[ChatCompletionReasoningItem] | None
) -> Delta:
return Delta(
content=message.get("content"),
role=message["role"],
function_call=message.get("function_call"),
tool_calls=message.get("tool_calls"),
reasoning_items=None if reasoning_items is None else list(reasoning_items),
provider_specific_fields=_cached_provider_specific_fields(
_CACHED_MESSAGE_PROVIDER_FIELDS.validate_python(message)
),
def _first_replay_delta(message: Mapping[str, object]) -> Delta:
return Delta.model_validate(
{
**message,
"provider_specific_fields": _cached_provider_specific_fields(
_CACHED_MESSAGE_PROVIDER_FIELDS.validate_python(message)
),
}
)
@ -234,7 +229,7 @@ async def convert_to_streaming_response_async(
t["index"] = index
pydantic_tool_calls.append(ChatCompletionDeltaToolCall(**t))
choice["message"]["tool_calls"] = pydantic_tool_calls
delta = _first_replay_delta(choice["message"], choice["message"].get("reasoning_items"))
delta = _first_replay_delta(choice["message"])
finish_reason = choice.get("finish_reason", None)
if finish_reason is None:

View file

@ -2413,6 +2413,75 @@ class TestConvertToStreamingResponseAsync:
assert chunks[-1].choices[0].finish_reason == "stop"
assert chunks[-1].usage.prompt_tokens == 3
def test_preserves_reasoning_fields(self):
import asyncio
from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (
convert_to_streaming_response_async,
)
thinking_blocks = [
{
"type": "thinking",
"thinking": "cached reasoning",
"signature": "sig-cache",
}
]
annotations = [
{
"type": "url_citation",
"url_citation": {
"start_index": 0,
"end_index": 5,
"title": "Cached source",
"url": "https://example.com/cached-source",
},
}
]
response_object = {
"id": "msg_async_reasoning_cache",
"model": "claude-3",
"created": 1700000000,
"choices": [
{
"finish_reason": "stop",
"index": 0,
"message": {
"content": "Final answer",
"role": "assistant",
"reasoning_content": "cached reasoning",
"thinking_blocks": thinking_blocks,
"annotations": annotations,
},
}
],
}
async def run():
return [
chunk
async for chunk in convert_to_streaming_response_async(
response_object=response_object
)
]
chunks = asyncio.run(run())
assert (
"".join(c.choices[0].delta.content or "" for c in chunks) == "Final answer"
)
assert chunks[0].choices[0].delta.reasoning_content == "cached reasoning"
assert chunks[0].choices[0].delta.thinking_blocks == thinking_blocks
assert chunks[0].choices[0].delta.annotations == annotations
assert not any(c.choices[0].delta.get("annotations") for c in chunks[1:])
assert not any(
hasattr(c.choices[0].delta, "reasoning_content") for c in chunks[1:]
)
assert not any(
hasattr(c.choices[0].delta, "thinking_blocks") for c in chunks[1:]
)
class TestHandleInvalidParallelToolCalls:
def test_none_input(self):
from litellm.litellm_core_utils.llm_response_utils.convert_dict_to_response import (