From 2647c2890fc6b8a32ab5601ecd8ed04b1cc0ebff Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Thu, 3 Sep 2026 01:25:32 -0700
Subject: [PATCH] fix(oci): keep the stream's own finish reason instead of a
 synthetic stop chunk

The OCI wrapper overrides chunk_creator wholesale, so it never recorded the finish reason or marked the terminal chunk as sent. The shared end-of-stream finalizer then appended a synthetic chunk whose finish_reason was always stop, which downgraded a tool_calls completion for any client that reads the finish reason off the last chunk.
---
 litellm/llms/oci/chat/transformation.py       | 12 +++--
 .../oci/chat/test_oci_chat_transformation.py  | 54 +++++++++++++++++++
 2 files changed, 63 insertions(+), 3 deletions(-)

diff --git a/litellm/llms/oci/chat/transformation.py b/litellm/llms/oci/chat/transformation.py
index b1167d447fc..8e4e41b4ac1 100644
--- a/litellm/llms/oci/chat/transformation.py
+++ b/litellm/llms/oci/chat/transformation.py
@@ -745,7 +745,13 @@ class OCIStreamWrapper(CustomStreamWrapper):
         # single-event case (terminal chunk carries the only copy of the text).
         self._cohere_text_emitted = False
 
-    def _with_stream_identity(self, parsed: ModelResponseStream) -> ModelResponseStream:
+    def _emit_chunk(self, parsed: ModelResponseStream) -> ModelResponseStream:
+        for choice in parsed.choices:
+            if getattr(choice.delta, "tool_calls", None):
+                self.tool_call = True
+            if choice.finish_reason is not None:
+                self.received_finish_reason = choice.finish_reason
+                self.sent_last_chunk = True
         return self.model_response_creator(chunk={"choices": parsed.choices})
 
     def chunk_creator(self, chunk: Any) -> ModelResponseStream | None:
@@ -780,8 +786,8 @@ class OCIStreamWrapper(CustomStreamWrapper):
                     if getattr(choice.delta, "content", None):
                         self._cohere_text_emitted = True
                         break
-            return self._with_stream_identity(result)
-        return self._with_stream_identity(handle_generic_stream_chunk(dict_chunk))
+            return self._emit_chunk(result)
+        return self._emit_chunk(handle_generic_stream_chunk(dict_chunk))
 
 
 __all__ = [
diff --git a/tests/test_litellm/llms/oci/chat/test_oci_chat_transformation.py b/tests/test_litellm/llms/oci/chat/test_oci_chat_transformation.py
index e77b2c24d01..4c9bd29b337 100644
--- a/tests/test_litellm/llms/oci/chat/test_oci_chat_transformation.py
+++ b/tests/test_litellm/llms/oci/chat/test_oci_chat_transformation.py
@@ -2041,3 +2041,57 @@ class TestOCIStreamWrapperDoneSentinel:
         wrapper = _make_stream_wrapper(_STREAM_GENERIC_MODEL)
         with pytest.raises(OCIError, match="Chunk cannot be parsed as JSON"):
             wrapper.chunk_creator("data: [DONE] trailing garbage")
+
+
+_GENERIC_TOOL_CALL_EVENT = (
+    'data: {"index":0,"message":{"role":"ASSISTANT","content":[],'
+    '"toolCalls":[{"type":"FUNCTION","id":"call_1","name":"get_weather","arguments":"{}"}]}}'
+)
+_GENERIC_TOOL_TERMINAL_EVENT = (
+    'data: {"index":0,"message":{"role":"ASSISTANT","content":[]},"finishReason":"TOOL_CALLS"}'
+)
+
+
+def _drain_stream(model: str, events: list[str]) -> list:
+    logging_obj = MagicMock()
+    logging_obj.model_call_details = {"custom_llm_provider": "oci", "litellm_params": {}}
+    wrapper = OCIStreamWrapper(
+        completion_stream=iter(events),
+        model=model,
+        custom_llm_provider="oci",
+        logging_obj=logging_obj,
+    )
+    return list(wrapper)
+
+
+class TestOCIStreamWrapperTerminalChunk:
+    """OCI's ``chunk_creator`` override bypasses the shared handler's
+    finish-reason bookkeeping, so the shared end-of-stream finalizer used to
+    append a synthetic ``stop`` chunk after OCI's own terminal chunk, silently
+    downgrading a ``tool_calls`` completion for any client that reads the
+    finish reason off the last chunk."""
+
+    def test_generic_tool_call_stream_ends_on_tool_calls(self):
+        chunks = _drain_stream(
+            _STREAM_GENERIC_MODEL,
+            [_GENERIC_TOOL_CALL_EVENT, _GENERIC_TOOL_TERMINAL_EVENT, "data: [DONE]"],
+        )
+
+        assert [chunk.choices[0].finish_reason for chunk in chunks] == [None, "tool_calls"]
+        assert len({chunk.id for chunk in chunks}) == 1
+
+    def test_generic_text_stream_emits_exactly_one_finish_reason(self):
+        chunks = _drain_stream(
+            _STREAM_GENERIC_MODEL,
+            [_GENERIC_TEXT_EVENT.format(text="1"), _GENERIC_TERMINAL_EVENT, "data: [DONE]"],
+        )
+
+        assert [chunk.choices[0].finish_reason for chunk in chunks] == [None, "stop"]
+
+    def test_cohere_stream_emits_exactly_one_finish_reason(self):
+        chunks = _drain_stream(
+            _STREAM_COHERE_MODEL,
+            [_COHERE_TEXT_EVENT.format(text="123"), _COHERE_TERMINAL_EVENT],
+        )
+
+        assert [chunk.choices[0].finish_reason for chunk in chunks] == [None, "stop"]