fix(base_utils): keep a closing developer message in place only after an assistant turn

A developer message that closes the conversation after a user or tool message hoists like
any other, so the request shape the ticket reports keeps working on system-first templates;
only a closing run that follows an assistant turn stays in place

litellm/llms/base_llm/base_utils.py

@@ -312,14 +312,17 @@ def _leading_system_block_length(messages: Sequence[AllMessageValues]) -> int:
 
 
 def _closing_instruction_block_start(messages: Sequence[AllMessageValues], leading_length: int) -> int:
-    return next(
+    last_conversation_index: Final = next(
         (
-            index + 1
+            index
             for index in range(len(messages) - 1, leading_length - 1, -1)
             if messages[index]["role"] not in _INSTRUCTION_ROLES
         ),
-        leading_length,
+        None,
     )
+    if last_conversation_index is None or messages[last_conversation_index]["role"] != "assistant":
+        return len(messages)
+    return last_conversation_index + 1
 
 
 def _move_later_developer_messages_up(messages: Sequence[AllMessageValues]) -> tuple[AllMessageValues, ...]:
@@ -347,9 +350,10 @@ def hoist_developer_messages_into_leading_system_message(
     Translate `developer` role to `system` role for OpenAI-compatible backends whose
     chat template allows a single system message and only at the start: developer
     messages that arrive after the first user turn move into the leading system
-    block, except a developer message that closes the conversation, which stays in
-    place so the request does not end on the assistant's turn. Each run of
-    consecutive system messages is then folded into one message in a single pass.
+    block, except a developer message that closes the conversation right after an
+    assistant turn, which stays in place so the request does not end on the
+    assistant's turn. Each run of consecutive system messages is then folded into
+    one message in a single pass.
     """
     translated: Final = tuple(map(_as_system_message, _move_later_developer_messages_up(messages)))
     return tuple(_merged_system_runs(translated))

tests/test_litellm/llms/base_llm/test_base_utils.py

@@ -275,12 +275,15 @@ class TestHoistDeveloperMessagesIntoLeadingSystemMessage:
             {"role": "system", "content": "Reply with the single word PONG and nothing else."},
         ]
 
-    def test_only_the_closing_developer_run_stays_in_place_and_is_folded_into_one_message(self):
+    def test_only_the_closing_developer_run_after_an_assistant_turn_stays_in_place_and_is_folded_into_one_message(
+        self,
+    ):
         messages = [
             {"role": "system", "content": "Base"},
             {"role": "user", "content": "Turn 1"},
             {"role": "developer", "content": "Update A"},
             {"role": "user", "content": "Turn 2"},
+            {"role": "assistant", "content": "Reply 2"},
             {"role": "developer", "content": "Closing B"},
             {"role": "developer", "content": "Closing C"},
         ]
@@ -288,9 +291,43 @@ class TestHoistDeveloperMessagesIntoLeadingSystemMessage:
             {"role": "system", "content": "Base\n\nUpdate A"},
             {"role": "user", "content": "Turn 1"},
             {"role": "user", "content": "Turn 2"},
+            {"role": "assistant", "content": "Reply 2"},
             {"role": "system", "content": "Closing B\n\nClosing C"},
         ]
 
+    def test_developer_message_that_closes_the_conversation_after_a_user_turn_is_hoisted(self):
+        messages = [
+            {"role": "user", "content": "Hi there"},
+            {"role": "developer", "content": "Reply with exactly one word: the capital of France"},
+        ]
+        assert list(hoist_developer_messages_into_leading_system_message(messages)) == [
+            {"role": "system", "content": "Reply with exactly one word: the capital of France"},
+            {"role": "user", "content": "Hi there"},
+        ]
+
+    def test_developer_message_that_closes_the_conversation_after_a_tool_result_is_hoisted(self):
+        messages = [
+            {"role": "system", "content": "Base"},
+            {"role": "user", "content": "Look it up"},
+            {
+                "role": "assistant",
+                "content": None,
+                "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "lookup", "arguments": "{}"}}],
+            },
+            {"role": "tool", "tool_call_id": "call_1", "content": "Paris"},
+            {"role": "developer", "content": "Answer with exactly one word."},
+        ]
+        assert list(hoist_developer_messages_into_leading_system_message(messages)) == [
+            {"role": "system", "content": "Base\n\nAnswer with exactly one word."},
+            {"role": "user", "content": "Look it up"},
+            {
+                "role": "assistant",
+                "content": None,
+                "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "lookup", "arguments": "{}"}}],
+            },
+            {"role": "tool", "tool_call_id": "call_1", "content": "Paris"},
+        ]
+
     def test_hoisted_block_content_developer_message_merges_as_blocks_after_string_instructions(self):
         messages = [
             {"role": "system", "content": "You are Codex"},