From 75c4565dde1ed63ecd8a3ef67ed1829da4888759 Mon Sep 17 00:00:00 2001
From: Deepanshu Lulla <deepanshu.lulla@gmail.com>
Date: Tue, 25 Aug 2026 17:10:56 -0400
Subject: [PATCH] fix(cerebras): add max_retries and extra_headers to
 get_supported_openai_params (#36601)

Co-authored-by: Deepanshu <deepanshu.lulla@alpha-sense.com>
---
 litellm/llms/cerebras/chat.py                 |  2 +
 tests/test_litellm/llms/cerebras/__init__.py  |  0
 .../test_cerebras_chat_transformation.py      | 61 +++++++++++++++++++
 3 files changed, 63 insertions(+)
 create mode 100644 tests/test_litellm/llms/cerebras/__init__.py
 create mode 100644 tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py

diff --git a/litellm/llms/cerebras/chat.py b/litellm/llms/cerebras/chat.py
index 8827b0afd87..c3aa26ade35 100644
--- a/litellm/llms/cerebras/chat.py
+++ b/litellm/llms/cerebras/chat.py
@@ -68,6 +68,8 @@ class CerebrasConfig(OpenAIGPTConfig):
             "tool_choice",
             "tools",
             "user",
+            "max_retries",
+            "extra_headers",
         ]
 
         # Only add reasoning_effort for models that support it
diff --git a/tests/test_litellm/llms/cerebras/__init__.py b/tests/test_litellm/llms/cerebras/__init__.py
new file mode 100644
index 00000000000..e69de29bb2d
diff --git a/tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py b/tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py
new file mode 100644
index 00000000000..09718b1e6e0
--- /dev/null
+++ b/tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py
@@ -0,0 +1,61 @@
+from litellm.llms.cerebras.chat import CerebrasConfig
+
+
+def test_max_retries_in_supported_params() -> None:
+    config = CerebrasConfig()
+    params = config.get_supported_openai_params(model="llama-3.3-70b")
+    assert "max_retries" in params, (
+        f"max_retries must be in CerebrasConfig.get_supported_openai_params(); got: {params!r}"
+    )
+
+
+def test_extra_headers_in_supported_params() -> None:
+    config = CerebrasConfig()
+    params = config.get_supported_openai_params(model="llama-3.3-70b")
+    assert "extra_headers" in params, (
+        f"extra_headers must be in CerebrasConfig.get_supported_openai_params(); got: {params!r}"
+    )
+
+
+def test_core_openai_params_still_supported() -> None:
+    config = CerebrasConfig()
+    params = config.get_supported_openai_params(model="llama-3.3-70b")
+    for expected in (
+        "max_tokens",
+        "max_completion_tokens",
+        "response_format",
+        "seed",
+        "stop",
+        "stream",
+        "temperature",
+        "top_p",
+        "tool_choice",
+        "tools",
+        "user",
+    ):
+        assert expected in params, f"{expected!r} unexpectedly missing from Cerebras supported params: {params!r}"
+
+
+def test_map_openai_params_preserves_max_retries() -> None:
+    config = CerebrasConfig()
+    result = config.map_openai_params(
+        non_default_params={"max_retries": 0, "temperature": 0.7},
+        optional_params={},
+        model="llama-3.3-70b",
+        drop_params=False,
+    )
+    assert result.get("max_retries") == 0, f"map_openai_params must preserve max_retries=0; got: {result!r}"
+    assert result.get("temperature") == 0.7
+
+
+def test_map_openai_params_preserves_max_retries_zero_falsy() -> None:
+    config = CerebrasConfig()
+    result = config.map_openai_params(
+        non_default_params={"max_retries": 0},
+        optional_params={},
+        model="llama-3.3-70b",
+        drop_params=False,
+    )
+    assert "max_retries" in result and result["max_retries"] == 0, (
+        f"max_retries=0 (falsy) must not be silently omitted; got: {result!r}"
+    )