From 42a80d5ade1e3ae832ec740f6794bde0e55dae98 Mon Sep 17 00:00:00 2001
From: Abdullah Alqahtani <61029571+anqorithm@users.noreply.github.com>
Date: Tue, 3 Nov 2026 01:47:49 +0100
Subject: [PATCH] fix(ollama): send generate keep_alive at the request root

---
 .../llms/ollama/completion/transformation.py  |  4 +++
 .../test_ollama_completion_transformation.py  | 25 +++++++++++++++++++
 2 files changed, 29 insertions(+)

diff --git a/litellm/llms/ollama/completion/transformation.py b/litellm/llms/ollama/completion/transformation.py
index 6dbcb050e11..430babd507c 100644
--- a/litellm/llms/ollama/completion/transformation.py
+++ b/litellm/llms/ollama/completion/transformation.py
@@ -462,6 +462,10 @@ class OllamaConfig(BaseConfig):
             "stream": stream,
         }
 
+        if "keep_alive" in optional_params:
+            if optional_params["keep_alive"] is not None:
+                data["keep_alive"] = optional_params["keep_alive"]
+            del optional_params["keep_alive"]  # rebind-ok: remove the promoted field from outgoing generation options
         if format is not None:
             data["format"] = format
         if images is not None:
diff --git a/tests/unit/llms/ollama/test_ollama_completion_transformation.py b/tests/unit/llms/ollama/test_ollama_completion_transformation.py
index 26addf1b24c..c65bae956bc 100644
--- a/tests/unit/llms/ollama/test_ollama_completion_transformation.py
+++ b/tests/unit/llms/ollama/test_ollama_completion_transformation.py
@@ -1100,3 +1100,28 @@ def test_reasoning_effort_ollama_honors_does_not_warn(
 
     assert result["think"] == think
     assert _reasoning_effort_warnings(caplog) == ()
+
+
+@pytest.mark.parametrize("keep_alive", (0, "0", "5m", -1))
+def test_generate_request_sends_keep_alive_outside_generation_options(keep_alive: int | str) -> None:
+    request: Final = OllamaConfig().transform_request(
+        model="test-model",
+        messages=[{"role": "user", "content": "hello"}],
+        optional_params={"keep_alive": keep_alive, "temperature": 0.25},
+        litellm_params={"text_completion": True},
+        headers={},
+    )
+    assert request["keep_alive"] == keep_alive
+    assert request["options"] == {"temperature": 0.25}
+
+
+def test_generate_request_omits_an_unset_keep_alive() -> None:
+    request: Final = OllamaConfig().transform_request(
+        model="test-model",
+        messages=[{"role": "user", "content": "hello"}],
+        optional_params={"keep_alive": None, "temperature": 0.25},
+        litellm_params={"text_completion": True},
+        headers={},
+    )
+    assert "keep_alive" not in request
+    assert request["options"] == {"temperature": 0.25}