A high-throughput and memory-efficient inference and serving engine for LLMs
Treat null completion max_tokens like the default (#45491)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jun 13, 2026 at 04:34 UTC
9261dbbc557b6bdd6b4f176a61e8008f2e99f3ed
1 file changed
+8
vllm/entrypoints/openai/completion/protocol.py
+8
@@ -346,6 +346,14 @@ class CompletionRequest(OpenAIBaseModel):
346
thinking_token_budget=self.thinking_token_budget,
347
)
348
349
+ @model_validator(mode="before")
350
+ @classmethod
351
+ def normalize_null_max_tokens(cls, data):
352
+ if isinstance(data, dict) and data.get("max_tokens") is None:
353
+ data = data.copy()
354
+ data["max_tokens"] = cls.model_fields["max_tokens"].default
355
+ return data
356
+
357
@model_validator(mode="before")
358
@classmethod
359
def validate_response_format(cls, data):