mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-12 03:48:56 +00:00
Document why reasoning_content keeps its existing meaning
The reasoning_items accumulator now carries every item, while reasoning_content still holds the summary text of the most recent one. That is intentional: widening reasoning_content would change what every existing consumer of that field sees, which is a separate change from the reported bug (#43620, items being overwritten). Recorded at the accumulation site so the asymmetry reads as a decision rather than an oversight.
This commit is contained in:
parent
d2569073ad
commit
8dbba3c0ba
1 changed files with 6 additions and 0 deletions
|
|
@ -883,6 +883,12 @@ class LiteLLMResponsesTransformationHandler(CompletionTransformationBridge):
|
|||
summary_raw=item.summary,
|
||||
)
|
||||
pending_reasoning_items = (*pending_reasoning_items, built_reasoning_item)
|
||||
# ``reasoning_content`` deliberately keeps its pre-existing meaning —
|
||||
# the summary text of the most recent reasoning item — even though
|
||||
# ``reasoning_items`` now carries all of them. Widening it to join
|
||||
# every item's summary would change what every existing consumer of
|
||||
# ``reasoning_content`` sees, which is a different fix from "stop
|
||||
# dropping reasoning items" and belongs in its own change.
|
||||
reasoning_content = " ".join(s["text"] for s in built_reasoning_item["summary"] if s.get("text"))
|
||||
|
||||
elif isinstance(item, ResponseOutputMessage):
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue