Skip to content

Generate stable message IDs for Chat Completions streams#128

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

Open
jacobbabula wants to merge 1 commit into
base:main
Choose a base branch
from
+136 −12
Open

Generate stable message IDs for Chat Completions streams #128

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
33 changes: 21 additions & 12 deletions chatkit/agents.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -91,6 +91,8 @@
)
from .widgets import Markdown, Text, WidgetRoot

_AGENTS_SDK_FAKE_RESPONSES_ID = "__fake_id__"


class ClientToolCall(BaseModel):
"""
Expand DownExpand Up@@ -501,6 +503,7 @@ async def stream_agent_response(
thread = context.thread
queue_iterator = _AsyncQueueIterator(context._events)
produced_items = set()
synthetic_message_ids: dict[str, str] = {}
streaming_thought: None | StreamingThoughtTracker = None
# item_id -> content_index -> annotation count
item_annotation_count: defaultdict[str, defaultdict[int, int]] = defaultdict(
Expand DownExpand Up@@ -592,23 +595,23 @@ def end_workflow(item: WorkflowItem):
continue
content = await _convert_content(event.part, converter)
yield ThreadItemUpdatedEvent(
item_id=event.item_id,
item_id=synthetic_message_ids.get(event.item_id, event.item_id),
update=AssistantMessageContentPartAdded(
content_index=event.content_index,
content=content,
),
)
elif event.type == "response.output_text.delta":
yield ThreadItemUpdatedEvent(
item_id=event.item_id,
item_id=synthetic_message_ids.get(event.item_id, event.item_id),
update=AssistantMessageContentPartTextDelta(
content_index=event.content_index,
delta=event.delta,
),
)
elif event.type == "response.output_text.done":
yield ThreadItemUpdatedEvent(
item_id=event.item_id,
item_id=synthetic_message_ids.get(event.item_id, event.item_id),
update=AssistantMessageContentPartDone(
content_index=event.content_index,
content=AssistantMessageContent(
Expand All@@ -620,16 +623,17 @@ def end_workflow(item: WorkflowItem):
elif event.type == "response.output_text.annotation.added":
annotation = await _convert_annotation(event.annotation, converter)
if annotation:
item_id = synthetic_message_ids.get(event.item_id, event.item_id)
# Manually track annotation indices per content part in case we drop an annotation that
# we can't convert to our internal representation (e.g. missing filename).
annotation_index = item_annotation_count[event.item_id][
annotation_index = item_annotation_count[item_id][
event.content_index
]
item_annotation_count[event.item_id][event.content_index] = (
item_annotation_count[item_id][event.content_index] = (
annotation_index + 1
)
yield ThreadItemUpdatedEvent(
item_id=event.item_id,
item_id=item_id,
update=AssistantMessageContentPartAnnotationAdded(
content_index=event.content_index,
annotation_index=annotation_index,
Expand All@@ -651,11 +655,14 @@ def end_workflow(item: WorkflowItem):
if item.type == "message":
if ctx.workflow_item:
yield end_workflow(ctx.workflow_item)
produced_items.add(item.id)
item_id = item.id
if item_id == _AGENTS_SDK_FAKE_RESPONSES_ID:
item_id = ctx.generate_id("message")
synthetic_message_ids[item.id] = item_id
produced_items.add(item_id)
yield ThreadItemAddedEvent(
item=AssistantMessageItem(
# Reusing the Responses message ID
id=item.id,
id=item_id,
thread_id=thread.id,
content=[
await _convert_content(c, converter)
Expand DownExpand Up@@ -763,11 +770,13 @@ def end_workflow(item: WorkflowItem):
elif event.type == "response.output_item.done":
item = event.item
if item.type == "message":
produced_items.add(item.id)
item_id = synthetic_message_ids.pop(item.id, item.id)
if item_id == _AGENTS_SDK_FAKE_RESPONSES_ID:
item_id = ctx.generate_id("message")
produced_items.add(item_id)
yield ThreadItemDoneEvent(
item=AssistantMessageItem(
# Reusing the Responses message ID
id=item.id,
id=item_id,
thread_id=thread.id,
content=[
await _convert_content(c, converter)
Expand Down
115 changes: 115 additions & 0 deletions tests/test_agents.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -922,6 +922,121 @@ async def test_stream_agent_response_maps_events():
assert future.done() is True


async def test_stream_agent_response_replaces_synthetic_chat_completion_ids():
generated_ids = iter(["message_1", "message_2"])
store = Mock()
store.generate_item_id = lambda item_type, thread, context: next(generated_ids)
store.load_thread_items = AsyncMock(return_value=Page())
store.add_thread_item = AsyncMock()
context = AgentContext(
previous_response_id=None, thread=thread, store=store, request_context=None
)
result = make_result()

for sequence_number, text in enumerate(["First", "Second"]):
result.add_event(
RawResponsesStreamEvent(
type="raw_response_event",
data=ResponseOutputItemAddedEvent(
type="response.output_item.added",
item=ResponseOutputMessage(
id="__fake_id__",
content=[],
role="assistant",
status="in_progress",
type="message",
),
output_index=0,
sequence_number=sequence_number * 5,
),
)
)
result.add_event(
RawResponsesStreamEvent(
type="raw_response_event",
data=ResponseContentPartAddedEvent(
type="response.content_part.added",
part=ResponseOutputText(
type="output_text",
text="",
annotations=[],
),
content_index=0,
item_id="__fake_id__",
output_index=0,
sequence_number=sequence_number * 5 + 1,
),
)
)
result.add_event(
RawResponsesStreamEvent(
type="raw_response_event",
data=ResponseTextDeltaEvent(
type="response.output_text.delta",
delta=text,
content_index=0,
item_id="__fake_id__",
logprobs=[],
output_index=0,
sequence_number=sequence_number * 5 + 2,
),
)
)
result.add_event(
RawResponsesStreamEvent(
type="raw_response_event",
data=ResponseTextDoneEvent(
type="response.output_text.done",
text=text,
content_index=0,
item_id="__fake_id__",
logprobs=[],
output_index=0,
sequence_number=sequence_number * 5 + 3,
),
)
)
result.add_event(
RawResponsesStreamEvent(
type="raw_response_event",
data=ResponseOutputItemDoneEvent(
type="response.output_item.done",
item=ResponseOutputMessage(
id="__fake_id__",
content=[
ResponseOutputText(
annotations=[], type="output_text", text=text
)
],
role="assistant",
status="completed",
type="message",
),
output_index=0,
sequence_number=sequence_number * 5 + 4,
),
)
)

result.done()
events = await all_events(stream_agent_response(context, result))

added = [event for event in events if isinstance(event, ThreadItemAddedEvent)]
updated = [event for event in events if isinstance(event, ThreadItemUpdatedEvent)]
done = [event for event in events if isinstance(event, ThreadItemDoneEvent)]

assert [event.item.id for event in added] == ["message_1", "message_2"]
assert [event.item_id for event in updated] == [
"message_1",
"message_1",
"message_1",
"message_2",
"message_2",
"message_2",
]
assert [event.item.id for event in done] == ["message_1", "message_2"]


@pytest.mark.parametrize(
"raw_event,expected_event",
[
Expand Down