Skip to content

fix streaming thinking/tool calling with anthropic #20077

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Oct 13, 2025
+82 −45
Merged

fix streaming thinking/tool calling with anthropic #20077

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
125 changes: 81 additions & 44 deletions llama-index-integrations/llms/llama-index-llms-anthropic/llama_index/llms/anthropic/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,6 +52,9 @@
ContentBlockStopEvent,
CitationsSearchResultLocation,
InputJSONDelta,
RawContentBlockDeltaEvent,
RawContentBlockStartEvent,
RawContentBlockStopEvent,
TextBlock,
TextDelta,
ThinkingBlock,
Expand Down Expand Up @@ -443,7 +446,8 @@ def stream_chat(
)

def gen() -> Generator[AnthropicChatResponse, None, None]:
content = [LITextBlock(text="")]
content = []
cur_block = None
content_delta = ""
thinking = None
cur_tool_calls: List[ToolUseBlock] = []
Expand All @@ -453,13 +457,13 @@ def gen() -> Generator[AnthropicChatResponse, None, None]:
tracked_citations: Set[str] = set()
role = MessageRole.ASSISTANT
for r in response:
if isinstance(r, ContentBlockDeltaEvent):
if isinstance(r, (ContentBlockDeltaEvent, RawContentBlockDeltaEvent)):
if isinstance(r.delta, TextDelta):
content_delta = r.delta.text or ""
if not isinstance(content[-1], LITextBlock):
content.append(LITextBlock(text=content_delta))
if not isinstance(cur_block, LITextBlock):
cur_block = LITextBlock(text=content_delta)
else:
content[-1].text += content_delta
cur_block.text += content_delta

elif isinstance(r.delta, CitationsDelta) and isinstance(
r.delta.citation, CitationsSearchResultLocation
Expand All @@ -480,23 +484,23 @@ def gen() -> Generator[AnthropicChatResponse, None, None]:
)
)
elif isinstance(r.delta, SignatureDelta):
if thinking is None:
thinking = ThinkingBlock(
signature=r.delta.signature,
thinking="",
type="thinking",
if not isinstance(cur_block, LIThinkingBlock):
cur_block = LIThinkingBlock(
content="",
additional_information={"signature": r.delta.signature},
)
else:
thinking.signature += r.delta.signature
cur_block.additional_information["signature"] += (
r.delta.signature
)
elif isinstance(r.delta, ThinkingDelta):
if thinking is None:
thinking = ThinkingBlock(
signature="",
thinking=r.delta.thinking,
type="thinking",
if cur_block is None:
cur_block = LIThinkingBlock(
content=r.delta.thinking or "",
additional_information={"signature": ""},
)
else:
thinking.thinking += r.delta.thinking
cur_block.content += r.delta.thinking
elif isinstance(r.delta, CitationsDelta):
# TODO: handle citation deltas
cur_citations.append(r.delta.citation.model_dump())
Expand All @@ -523,29 +527,44 @@ def gen() -> Generator[AnthropicChatResponse, None, None]:
tool_calls_to_send = [*cur_tool_calls, cur_tool_call]
else:
tool_calls_to_send = cur_tool_calls

yield AnthropicChatResponse(
message=ChatMessage(
role=role,
blocks=content,
additional_kwargs={
"tool_calls": [
t.model_dump() for t in tool_calls_to_send
],
"thinking": thinking.model_dump() if thinking else None,
"tool_calls": [t.dict() for t in tool_calls_to_send]
},
),
citations=cur_citations,
delta=content_delta,
raw=dict(r),
)
elif isinstance(r, ContentBlockStartEvent):
elif isinstance(r, (ContentBlockStartEvent, RawContentBlockStartEvent)):
if isinstance(r.content_block, ToolUseBlock):
cur_tool_call = r.content_block
cur_tool_json = ""
elif isinstance(r, ContentBlockStopEvent):
elif isinstance(r, (ContentBlockStopEvent, RawContentBlockStopEvent)):
if isinstance(cur_tool_call, ToolUseBlock):
cur_tool_calls.append(cur_tool_call)

if cur_block is not None:
content.append(cur_block)
cur_block = None

yield AnthropicChatResponse(
message=ChatMessage(
role=role,
blocks=content,
additional_kwargs={
"tool_calls": [t.dict() for t in tool_calls_to_send]
},
),
citations=cur_citations,
delta=content_delta,
raw=dict(r),
)

return gen()

@llm_completion_callback()
Expand Down Expand Up @@ -615,7 +634,8 @@ async def astream_chat(
)

async def gen() -> ChatResponseAsyncGen:
content = [LITextBlock(text="")]
content = []
cur_block = None
content_delta = ""
thinking = None
cur_tool_calls: List[ToolUseBlock] = []
Expand All @@ -625,13 +645,13 @@ async def gen() -> ChatResponseAsyncGen:
tracked_citations: Set[str] = set()
role = MessageRole.ASSISTANT
async for r in response:
if isinstance(r, ContentBlockDeltaEvent):
if isinstance(r, (ContentBlockDeltaEvent, RawContentBlockDeltaEvent)):
if isinstance(r.delta, TextDelta):
content_delta = r.delta.text or ""
if not isinstance(content[-1], LITextBlock):
content.append(LITextBlock(text=content_delta))
if not isinstance(cur_block, LITextBlock):
cur_block = LITextBlock(text=content_delta)
else:
content[-1].text += content_delta
cur_block.text += content_delta

elif isinstance(r.delta, CitationsDelta) and isinstance(
r.delta.citation, CitationsSearchResultLocation
Expand All @@ -652,23 +672,23 @@ async def gen() -> ChatResponseAsyncGen:
)
)
elif isinstance(r.delta, SignatureDelta):
if thinking is None:
thinking = ThinkingBlock(
signature=r.delta.signature,
thinking="",
type="thinking",
if not isinstance(cur_block, LIThinkingBlock):
cur_block = LIThinkingBlock(
content="",
additional_information={"signature": r.delta.signature},
)
else:
thinking.signature += r.delta.signature
cur_block.additional_information["signature"] += (
r.delta.signature
)
elif isinstance(r.delta, ThinkingDelta):
if thinking is None:
thinking = ThinkingBlock(
signature="",
thinking=r.delta.thinking,
type="thinking",
if cur_block is None:
cur_block = LIThinkingBlock(
content=r.delta.thinking or "",
additional_information={"signature": ""},
)
else:
thinking.thinking += r.delta.thinking
cur_block.content += r.delta.thinking
elif isinstance(r.delta, CitationsDelta):
# TODO: handle citation deltas
cur_citations.append(r.delta.citation.model_dump())
Expand All @@ -695,27 +715,44 @@ async def gen() -> ChatResponseAsyncGen:
tool_calls_to_send = [*cur_tool_calls, cur_tool_call]
else:
tool_calls_to_send = cur_tool_calls

yield AnthropicChatResponse(
message=ChatMessage(
role=role,
blocks=content,
additional_kwargs={
"tool_calls": [t.dict() for t in tool_calls_to_send],
"thinking": thinking.model_dump() if thinking else None,
"tool_calls": [t.dict() for t in tool_calls_to_send]
},
),
citations=cur_citations,
delta=content_delta,
raw=dict(r),
)
elif isinstance(r, ContentBlockStartEvent):
elif isinstance(r, (ContentBlockStartEvent, RawContentBlockStartEvent)):
if isinstance(r.content_block, ToolUseBlock):
cur_tool_call = r.content_block
cur_tool_json = ""
elif isinstance(r, ContentBlockStopEvent):
elif isinstance(r, (ContentBlockStopEvent, RawContentBlockStopEvent)):
if isinstance(cur_tool_call, ToolUseBlock):
cur_tool_calls.append(cur_tool_call)

if cur_block is not None:
content.append(cur_block)
cur_block = None

yield AnthropicChatResponse(
message=ChatMessage(
role=role,
blocks=content,
additional_kwargs={
"tool_calls": [t.dict() for t in tool_calls_to_send]
},
),
citations=cur_citations,
delta=content_delta,
raw=dict(r),
)

return gen()

@llm_completion_callback()
Expand Down
2 changes: 1 addition & 1 deletion llama-index-integrations/llms/llama-index-llms-anthropic/pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ dev = [

[project]
name = "llama-index-llms-anthropic"
version = "0.9.3"
version = "0.9.4"
description = "llama-index llms anthropic integration"
authors = [{name = "Your Name", email = "[email protected]"}]
requires-python = ">=3.9,<4.0"
Expand Down
Loading