From 298a2fd0b975fcb8f365956600773b4bde5c4683 Mon Sep 17 00:00:00 2001
From: ly-wang19 <ly-wang19@users.noreply.github.com>
Date: Tue, 23 Jun 2026 18:41:44 +0800
Subject: [PATCH] fix(oci): only emit streaming finish_reason="tool_calls" on
 the terminal chunk
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Addresses @greptileai's P1. handle_generic_stream_chunk set
finish_reason="tool_calls" whenever tool_calls were present, including on
intermediate chunks — OCI streams tool calls progressively, so a non-terminal
chunk arrives with finishReason=null and toolCalls populated. Per OpenAI
streaming semantics finish_reason must be null on non-final deltas; a non-null
value there signals premature end-of-stream to callers that stop on
finish_reason is not None.

Guard the override with `finish_reason is not None` (matching the cohere
streaming handler). Adds a regression test for an intermediate chunk
(finishReason=null + toolCalls) asserting finish_reason stays None.
---
 litellm/llms/oci/chat/generic.py              |  8 +++++--
 .../llms/oci/chat/test_oci_finish_reason.py   | 23 +++++++++++++++++++
 2 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/litellm/llms/oci/chat/generic.py b/litellm/llms/oci/chat/generic.py
index 58555c30c6d..0394f5e001f 100644
--- a/litellm/llms/oci/chat/generic.py
+++ b/litellm/llms/oci/chat/generic.py
@@ -463,8 +463,12 @@ def handle_generic_stream_chunk(dict_chunk: dict) -> ModelResponseStream:
     finish_reason: Optional[str] = _normalize_oci_finish_reason(
         typed_chunk.finishReason
     )
-    # OpenAI semantics: a chunk carrying tool calls reports finish_reason="tool_calls".
-    if tool_calls:
+    # OpenAI semantics: the terminal chunk reports finish_reason="tool_calls" when
+    # tool calls are present. Guard with `finish_reason is not None` so intermediate
+    # chunks (OCI streams tool calls progressively, with finishReason=null and
+    # toolCalls populated) keep finish_reason=None — a non-null value on a
+    # non-terminal chunk would signal premature stream termination to callers.
+    if finish_reason is not None and tool_calls:
         finish_reason = "tool_calls"
 
     return ModelResponseStream(
diff --git a/tests/test_litellm/llms/oci/chat/test_oci_finish_reason.py b/tests/test_litellm/llms/oci/chat/test_oci_finish_reason.py
index 41b4b886ded..a8ac5419ab9 100644
--- a/tests/test_litellm/llms/oci/chat/test_oci_finish_reason.py
+++ b/tests/test_litellm/llms/oci/chat/test_oci_finish_reason.py
@@ -145,3 +145,26 @@ def test_generic_stream_chunk_with_tool_calls_reports_tool_calls():
     }
     result = handle_generic_stream_chunk(chunk)
     assert result.choices[0].finish_reason == "tool_calls"
+
+
+def test_generic_stream_intermediate_chunk_keeps_finish_reason_none():
+    # OCI streams tool calls progressively: an intermediate chunk carries
+    # toolCalls but finishReason=null. finish_reason must stay None on a
+    # non-terminal chunk — emitting "tool_calls" here would signal premature
+    # stream termination to callers that stop on finish_reason is not None.
+    chunk = {
+        "finishReason": None,
+        "message": {
+            "toolCalls": [
+                {
+                    "id": "call_0",
+                    "type": "FUNCTION",
+                    "name": "get_weather",
+                    "arguments": "{}",
+                }
+            ]
+        },
+        "index": 0,
+    }
+    result = handle_generic_stream_chunk(chunk)
+    assert result.choices[0].finish_reason is None