fix(cerebras): include max_retries in supported OpenAI params allowlist

``max_retries`` is a standard parameter exposed by both the OpenAI Python
client and LiteLLM's ``litellm.completion()`` API. The Cerebras inference
endpoint accepts it without issue, but ``CerebrasConfig.get_supported_openai_params``
was omitting it from the allowlist, so LiteLLM was silently stripping the
value before forwarding the call. Callers passing ``max_retries=N`` on
Cerebras requests therefore had no way to override the default retry
behaviour — the request flowed through unchanged but the override was
dropped without warning.

Adding ``"max_retries"`` to the allowlist lets the value reach the
underlying HTTP client and lets per-call retry overrides work as
documented.

Tests (tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py):
- test_max_retries_in_supported_params — guards against future
  regressions removing the entry.
- test_core_openai_params_still_supported — regression guard ensuring
  no existing allowlist entries are accidentally dropped by the change.

litellm/llms/cerebras/chat.py

@@ -68,6 +68,7 @@ class CerebrasConfig(OpenAIGPTConfig):
             "tool_choice",
             "tools",
             "user",
+            "max_retries",
         ]
 
         # Only add reasoning_effort for models that support it

tests/test_litellm/llms/cerebras/test_cerebras_chat_transformation.py

@@ -0,0 +1,53 @@
+"""
+Unit tests for Cerebras chat configuration.
+"""
+
+import os
+import sys
+
+sys.path.insert(
+    0, os.path.abspath("../../../../..")
+)  # project root for ``litellm`` import
+
+from litellm.llms.cerebras.chat import CerebrasConfig
+
+
+class TestCerebrasGetSupportedOpenAIParams:
+    """Validates the OpenAI-param allowlist exposed by ``CerebrasConfig``."""
+
+    def test_max_retries_in_supported_params(self):
+        """``max_retries`` is a standard OpenAI client parameter that Cerebras
+        accepts. It must appear in the supported-param allowlist so callers
+        can override the default retry count per request without LiteLLM
+        stripping the value silently.
+        """
+        config = CerebrasConfig()
+        params = config.get_supported_openai_params(model="llama3.1-8b")
+        assert "max_retries" in params, (
+            f"max_retries should be in the Cerebras supported_params allowlist; "
+            f"got: {params!r}"
+        )
+
+    def test_core_openai_params_still_supported(self):
+        """Regression guard: the standard OpenAI params Cerebras has always
+        accepted must remain in the allowlist after the ``max_retries``
+        addition (no accidental removals)."""
+        config = CerebrasConfig()
+        params = config.get_supported_openai_params(model="llama3.1-8b")
+        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: "
+                f"{params!r}"
+            )