mirror of
https://github.com/BerriAI/litellm.git
synced 2026-09-14 23:21:35 +00:00
fix(anthropic): skip non-data SSE lines in stream iterator to preserve streamed tool calls
This commit is contained in:
elyar1124
parent
ec3f8183c3
commit
efaca40574
2 changed files with 143 additions and 17 deletions
|
|
@ -47,7 +47,6 @@ from litellm.types.responses.main import (
|
|||
)
|
||||
from litellm.types.utils import (
|
||||
Delta,
|
||||
GenericStreamingChunk,
|
||||
LlmProviders,
|
||||
ModelResponse,
|
||||
ModelResponseStream,
|
||||
|
|
@ -1255,14 +1254,7 @@ class ModelResponseIterator:
|
|||
return result
|
||||
# If None, continue loop to get more chunks for accumulation
|
||||
else:
|
||||
return GenericStreamingChunk(
|
||||
text="",
|
||||
is_finished=False,
|
||||
finish_reason="",
|
||||
usage=None,
|
||||
index=0,
|
||||
tool_use=None,
|
||||
)
|
||||
continue
|
||||
except StopIteration:
|
||||
raise StopIteration
|
||||
except ValueError as e:
|
||||
|
|
@ -1301,14 +1293,7 @@ class ModelResponseIterator:
|
|||
return result
|
||||
# If None, continue loop to get more chunks for accumulation
|
||||
else:
|
||||
return GenericStreamingChunk(
|
||||
text="",
|
||||
is_finished=False,
|
||||
finish_reason="",
|
||||
usage=None,
|
||||
index=0,
|
||||
tool_use=None,
|
||||
)
|
||||
continue
|
||||
except StopAsyncIteration:
|
||||
raise StopAsyncIteration
|
||||
except ValueError as e:
|
||||
|
|
|
|||
|
|
@ -2568,3 +2568,144 @@ class TestRustChatCompletionsHook:
|
|||
"model": "m",
|
||||
"messages": [],
|
||||
}
|
||||
|
||||
|
||||
def test_anthropic_streaming_thinking_text_tool_use_sync():
|
||||
"""
|
||||
Regression test for Issue #36262:
|
||||
Ensure Anthropic SSE stream containing thinking -> text -> tool_use emits only
|
||||
ModelResponseStream instances (no dummy GenericStreamingChunk dicts),
|
||||
preserves delta.tool_calls, and successfully builds via stream_chunk_builder.
|
||||
"""
|
||||
sse_lines = [
|
||||
b'event: message_start\n',
|
||||
b'data: {"type":"message_start","message":{"id":"msg_test_123","type":"message","role":"assistant","model":"claude-3-7-sonnet-20250219","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":10,"output_tokens":1}}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_start\n',
|
||||
b'data: {"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_delta\n',
|
||||
b'data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"Let me call submit_result"}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_stop\n',
|
||||
b'data: {"type":"content_block_stop","index":0}\n',
|
||||
b'\n',
|
||||
b'event: content_block_start\n',
|
||||
b'data: {"type":"content_block_start","index":1,"content_block":{"type":"text","text":""}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_delta\n',
|
||||
b'data: {"type":"content_block_delta","index":1,"delta":{"type":"text_delta","text":"I will now submit the result."}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_stop\n',
|
||||
b'data: {"type":"content_block_stop","index":1}\n',
|
||||
b'\n',
|
||||
b'event: content_block_start\n',
|
||||
b'data: {"type":"content_block_start","index":2,"content_block":{"type":"tool_use","id":"toolu_01","name":"submit_result","input":{}}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_delta\n',
|
||||
b'data: {"type":"content_block_delta","index":2,"delta":{"type":"input_json_delta","partial_json":"{\\"value\\":"}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_delta\n',
|
||||
b'data: {"type":"content_block_delta","index":2,"delta":{"type":"input_json_delta","partial_json":"1}"}}\n',
|
||||
b'\n',
|
||||
b'event: content_block_stop\n',
|
||||
b'data: {"type":"content_block_stop","index":2}\n',
|
||||
b'\n',
|
||||
b'event: message_delta\n',
|
||||
b'data: {"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"output_tokens":25}}\n',
|
||||
b'\n',
|
||||
b'event: message_stop\n',
|
||||
b'data: {"type":"message_stop"}\n',
|
||||
]
|
||||
|
||||
iterator = ModelResponseIterator(
|
||||
streaming_response=iter(sse_lines),
|
||||
sync_stream=True,
|
||||
json_mode=False,
|
||||
)
|
||||
chunks = list(iterator)
|
||||
|
||||
# All emitted chunks must be ModelResponseStream instances
|
||||
assert all(isinstance(c, litellm.ModelResponseStream) for c in chunks)
|
||||
|
||||
# Tool call chunks must be present
|
||||
tool_call_chunks = [
|
||||
c for c in chunks
|
||||
if getattr(c.choices[0].delta, "tool_calls", None) is not None
|
||||
]
|
||||
assert len(tool_call_chunks) == 3
|
||||
assert tool_call_chunks[0].choices[0].delta.tool_calls[0].function.name == "submit_result"
|
||||
assert tool_call_chunks[0].choices[0].delta.tool_calls[0].id == "toolu_01"
|
||||
|
||||
# stream_chunk_builder must reconstruct the message with tool calls
|
||||
built = litellm.stream_chunk_builder(chunks=chunks)
|
||||
assert built.choices[0].message.content == "I will now submit the result."
|
||||
assert built.choices[0].message.tool_calls is not None
|
||||
assert len(built.choices[0].message.tool_calls) == 1
|
||||
assert built.choices[0].message.tool_calls[0].function.name == "submit_result"
|
||||
assert built.choices[0].message.tool_calls[0].function.arguments == '{"value":1}'
|
||||
assert built.choices[0].finish_reason == "tool_calls"
|
||||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_anthropic_streaming_thinking_text_tool_use_async():
|
||||
"""
|
||||
Regression test for Issue #36262 in async streaming (__anext__):
|
||||
Ensure async iteration correctly yields all ModelResponseStream chunks and preserves tool_calls.
|
||||
"""
|
||||
sse_lines = [
|
||||
'event: message_start\n',
|
||||
'data: {"type":"message_start","message":{"id":"msg_async_123","type":"message","role":"assistant","model":"claude-3-7-sonnet-20250219","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":10,"output_tokens":1}}}\n',
|
||||
'event: content_block_start\n',
|
||||
'data: {"type":"content_block_start","index":0,"content_block":{"type":"thinking","thinking":""}}\n',
|
||||
'event: content_block_delta\n',
|
||||
'data: {"type":"content_block_delta","index":0,"delta":{"type":"thinking_delta","thinking":"Async thinking..."}}\n',
|
||||
'event: content_block_stop\n',
|
||||
'data: {"type":"content_block_stop","index":0}\n',
|
||||
'event: content_block_start\n',
|
||||
'data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"toolu_async_01","name":"do_action","input":{}}}\n',
|
||||
'event: content_block_delta\n',
|
||||
'data: {"type":"content_block_delta","index":1,"delta":{"type":"input_json_delta","partial_json":"{\\"action\\": \\"run\\"}"}}\n',
|
||||
'event: content_block_stop\n',
|
||||
'data: {"type":"content_block_stop","index":1}\n',
|
||||
'event: message_delta\n',
|
||||
'data: {"type":"message_delta","delta":{"stop_reason":"tool_use","stop_sequence":null},"usage":{"output_tokens":15}}\n',
|
||||
'event: message_stop\n',
|
||||
'data: {"type":"message_stop"}\n',
|
||||
]
|
||||
|
||||
class AsyncLineIterator:
|
||||
def __init__(self, lines):
|
||||
self._iter = iter(lines)
|
||||
|
||||
def __aiter__(self):
|
||||
return self
|
||||
|
||||
async def __anext__(self):
|
||||
try:
|
||||
return next(self._iter)
|
||||
except StopIteration:
|
||||
raise StopAsyncIteration
|
||||
|
||||
iterator = ModelResponseIterator(
|
||||
streaming_response=AsyncLineIterator(sse_lines),
|
||||
sync_stream=False,
|
||||
json_mode=False,
|
||||
)
|
||||
|
||||
chunks = []
|
||||
async for chunk in iterator:
|
||||
chunks.append(chunk)
|
||||
|
||||
assert all(isinstance(c, litellm.ModelResponseStream) for c in chunks)
|
||||
tool_call_chunks = [
|
||||
c for c in chunks
|
||||
if getattr(c.choices[0].delta, "tool_calls", None) is not None
|
||||
]
|
||||
assert len(tool_call_chunks) == 2
|
||||
assert tool_call_chunks[0].choices[0].delta.tool_calls[0].function.name == "do_action"
|
||||
|
||||
built = litellm.stream_chunk_builder(chunks=chunks)
|
||||
assert built.choices[0].message.tool_calls[0].function.name == "do_action"
|
||||
assert json.loads(built.choices[0].message.tool_calls[0].function.arguments) == {"action": "run"}
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue