From 500e267b3f7d8fc5d0444ca5e9288d0fb6779bc8 Mon Sep 17 00:00:00 2001
From: Ramzi Alashmali <186906047+RAMZI0TO99@users.noreply.github.com>
Date: Tue, 6 Oct 2026 09:13:48 +0300
Subject: [PATCH] fix(openai): honor zero retries for embeddings

---
 litellm/llms/openai/openai.py         |  8 ++--
 tests/unit/llms/openai/test_openai.py | 59 +++++++++++++++++++++++++++
 2 files changed, 63 insertions(+), 4 deletions(-)

diff --git a/litellm/llms/openai/openai.py b/litellm/llms/openai/openai.py
index 2461fb4e1af..c9faff270e0 100644
--- a/litellm/llms/openai/openai.py
+++ b/litellm/llms/openai/openai.py
@@ -1332,8 +1332,8 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM):
         super().embedding()
         try:
             data: Final = {"model": model, "input": input, **optional_params}
-            max_retries = max_retries or litellm.DEFAULT_MAX_RETRIES
-            if not isinstance(max_retries, int):
+            effective_max_retries: Final = litellm.DEFAULT_MAX_RETRIES if max_retries is None else max_retries
+            if not isinstance(effective_max_retries, int):
                 raise OpenAIError(status_code=422, message="max retries must be an int")
             ## LOGGING
             logging_obj.pre_call(
@@ -1352,7 +1352,7 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM):
                     api_key=api_key,
                     timeout=timeout,
                     client=client,
-                    max_retries=max_retries,
+                    max_retries=effective_max_retries,
                     shared_session=shared_session,
                     litellm_params=litellm_params,
                 )
@@ -1362,7 +1362,7 @@ class OpenAIChatCompletion(BaseLLM, BaseOpenAILLM):
                 api_key=api_key,
                 api_base=api_base,
                 timeout=timeout,
-                max_retries=max_retries,
+                max_retries=effective_max_retries,
                 client=client,
                 litellm_params=litellm_params,
             )
diff --git a/tests/unit/llms/openai/test_openai.py b/tests/unit/llms/openai/test_openai.py
index 2be691fa65b..b4badf6e5bc 100644
--- a/tests/unit/llms/openai/test_openai.py
+++ b/tests/unit/llms/openai/test_openai.py
@@ -350,3 +350,62 @@ async def test_async_audio_speech_records_provider_response_headers():
         )
 
     _assert_provider_headers_recorded(response)
+
+
+def _embedding_transport() -> httpx.MockTransport:
+    return httpx.MockTransport(
+        lambda request: httpx.Response(
+            200,
+            json={
+                "object": "list",
+                "model": "embedding-test",
+                "data": [{"object": "embedding", "index": 0, "embedding": [0.1, 0.2]}],
+                "usage": {"prompt_tokens": 1, "total_tokens": 1},
+            },
+        )
+    )
+
+
+@pytest.mark.parametrize("max_retries,expected_retries", [(0, 0), (3, 3), (None, litellm.DEFAULT_MAX_RETRIES)])
+def test_embedding_honors_explicit_retry_limit(max_retries: int | None, expected_retries: int) -> None:
+    with OpenAI(
+        api_key="transport-only",
+        base_url="https://embedding.example/v1",
+        max_retries=7,
+        http_client=httpx.Client(transport=_embedding_transport()),
+    ) as client:
+        response: Final = litellm.embedding(
+            model="openai/embedding-test",
+            input=["retry control"],
+            encoding_format="float",
+            client=client,
+            max_retries=max_retries,
+            api_key="transport-only",
+            api_base=str(client.base_url),
+        )
+
+    assert client.max_retries == expected_retries
+    assert response.data == [{"object": "embedding", "index": 0, "embedding": [0.1, 0.2]}]
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("max_retries,expected_retries", [(0, 0), (3, 3), (None, litellm.DEFAULT_MAX_RETRIES)])
+async def test_aembedding_honors_explicit_retry_limit(max_retries: int | None, expected_retries: int) -> None:
+    async with AsyncOpenAI(
+        api_key="transport-only",
+        base_url="https://embedding.example/v1",
+        max_retries=7,
+        http_client=httpx.AsyncClient(transport=_embedding_transport()),
+    ) as client:
+        response: Final = await litellm.aembedding(
+            model="openai/embedding-test",
+            input=["retry control"],
+            encoding_format="float",
+            client=client,
+            max_retries=max_retries,
+            api_key="transport-only",
+            api_base=str(client.base_url),
+        )
+
+    assert client.max_retries == expected_retries
+    assert response.data == [{"object": "embedding", "index": 0, "embedding": [0.1, 0.2]}]