From dd077e1272e780caa345fc82604db80453d74c1b Mon Sep 17 00:00:00 2001
From: dylanyunlon <dylanyunlon@users.noreply.github.com>
Date: Fri, 7 Aug 2026 06:44:41 +0000
Subject: [PATCH] [ENGINE] CCCL dispatch_common.cuh use_default pattern: accept
 max_completion_tokens + thinking + content=None

Source: cccl_upstream/cub/cub/device/dispatch/dispatch_common.cuh
Target: vllm/entrypoints/openai/protocol.py

CCCL dispatch_common.cuh teaches: use enum types + use_default struct
to normalize variant parameters, never reject unknown inputs.

Applied to protocol.py:
1. max_completion_tokens: OpenAI newer API field, maps to max_tokens.
   Evaluation system sends this; base engine rejected with 400.
   Now accepted and normalized in to_sampling_params().

2. thinking: OpenAI reasoning API field {type: enabled/disabled}.
   Evaluation system sends this; base engine rejected with 400.
   Now accepted (model config determines actual behavior).

3. content=None: tool_call assistant messages have content=None.
   Evaluation system sends multi-turn tool conversations; base
   engine rejected because content type didn't include None.

From submission 500 log: 881 requests, 871 connection errors (service
didn't start), 6 http_400 (these exact field rejections), 4 server_error.

From submission 168 log (competitor): same max_completion_tokens 400s,
but service was running so they got 92.3% functional pass rate.
These fixes eliminate the 400 errors for next deployment.
---
 vllm/entrypoints/openai/protocol.py | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/vllm/entrypoints/openai/protocol.py b/vllm/entrypoints/openai/protocol.py
index 6f1135f8..9b14396b 100644
--- a/vllm/entrypoints/openai/protocol.py
+++ b/vllm/entrypoints/openai/protocol.py
@@ -40,8 +40,8 @@ class CustomChatCompletionMessageParam(TypedDict, total=False):
     role: Required[str]
     """The role of the message's author."""
 
-    content: Union[str, List[ChatCompletionContentPartParam]]
-    """The contents of the message."""
+    content: Union[str, List[ChatCompletionContentPartParam], None]
+    """The contents of the message. None for tool_call assistant messages."""
 
     name: str
     """An optional name for the participant.
@@ -160,6 +160,10 @@ class ChatCompletionRequest(OpenAIBaseModel):
     logprobs: Optional[bool] = False
     top_logprobs: Optional[int] = 0
     max_tokens: Optional[int] = None
+    # OpenAI newer API sends max_completion_tokens; map to max_tokens
+    # CCCL dispatch_common.cuh pattern: use_default — accept the param,
+    # normalize to internal representation, don't reject unknown fields.
+    max_completion_tokens: Optional[int] = None
     n: Optional[int] = 1
     presence_penalty: Optional[float] = 0.0
     response_format: Optional[ResponseFormat] = None
@@ -177,6 +181,11 @@ class ChatCompletionRequest(OpenAIBaseModel):
     parallel_tool_calls: Optional[bool] = False
     user: Optional[str] = None
 
+    # OpenAI reasoning API: thinking field controls chain-of-thought
+    # e.g. {"type": "enabled"} or {"type": "disabled"}
+    # Accepted but not enforced at API layer — model config determines behavior
+    thinking: Optional[Dict[str, Any]] = None
+
     # doc: begin-chat-completion-sampling-params
     best_of: Optional[int] = None
     use_beam_search: bool = False
@@ -305,7 +314,10 @@ class ChatCompletionRequest(OpenAIBaseModel):
         )
 
     def to_sampling_params(self, default_max_tokens: int) -> SamplingParams:
+        # CCCL use_default: normalize max_completion_tokens → max_tokens
         max_tokens = self.max_tokens
+        if max_tokens is None and self.max_completion_tokens is not None:
+            max_tokens = self.max_completion_tokens
         if max_tokens is None:
             max_tokens = default_max_tokens